blob: 56baae8c2f56baf0f41bb7bb24b0b12679267d20 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Paolo Bonzini2f303b72013-09-25 13:53:07 +020073DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020074static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080075LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080076
Rusty Russell7f59f492008-12-07 21:25:45 +103077static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020078static int kvm_usage_count = 0;
79static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030080
Rusty Russellc16f8622007-07-30 21:12:19 +100081struct kmem_cache *kvm_vcpu_cache;
82EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030083
Avi Kivity15ad7142007-07-11 18:17:21 +030084static __read_mostly struct preempt_ops kvm_preempt_ops;
85
Hollis Blanchard76f7c872008-04-15 16:05:42 -050086struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080087
Avi Kivitybccf2152007-02-21 18:04:26 +020088static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
89 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020090#ifdef CONFIG_COMPAT
91static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
92 unsigned long arg);
93#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020094static int hardware_enable_all(void);
95static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020096
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020097static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -080098static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
100
101static void kvm_release_pfn_dirty(pfn_t pfn);
102static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200104
Andi Kleen52480132014-02-08 08:51:57 +0100105__visible bool kvm_rebooting;
Avi Kivityb7c41452010-12-02 17:52:50 +0200106EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300107
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300108static bool largepages_enabled = true;
109
Xiao Guangronga2766322012-07-26 11:58:59 +0800110bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300111{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200112 if (pfn_valid(pfn))
113 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300114
115 return true;
116}
117
Avi Kivity6aa8b732006-12-10 02:21:36 -0800118/*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300121int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800122{
Avi Kivity15ad7142007-07-11 18:17:21 +0300123 int cpu;
124
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
132 synchronize_rcu();
133 put_pid(oldpid);
134 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300135 cpu = get_cpu();
136 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200137 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300138 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300139 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200140}
141
Carsten Otte313a3dc2007-10-11 19:16:52 +0200142void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800143{
Avi Kivity15ad7142007-07-11 18:17:21 +0300144 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200145 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300146 preempt_notifier_unregister(&vcpu->preempt_notifier);
147 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800148 mutex_unlock(&vcpu->mutex);
149}
150
Avi Kivityd9e368d2007-06-07 19:18:30 +0300151static void ack_flush(void *_completed)
152{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300153}
154
Rusty Russell49846892008-12-08 20:26:24 +1030155static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300156{
Avi Kivity597a5f52008-07-20 14:24:22 +0300157 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030158 cpumask_var_t cpus;
159 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300160 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300161
Li Zefan79f55992009-06-15 14:58:26 +0800162 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030163
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800164 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300165 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800166 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300167 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800168
169 /* Set ->requests bit before we read ->mode */
170 smp_mb();
171
172 if (cpus != NULL && cpu != -1 && cpu != me &&
173 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030174 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300175 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176 if (unlikely(cpus == NULL))
177 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
178 else if (!cpumask_empty(cpus))
179 smp_call_function_many(cpus, ack_flush, NULL, 1);
180 else
181 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800182 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030183 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030184 return called;
185}
186
187void kvm_flush_remote_tlbs(struct kvm *kvm)
188{
189 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
190 ++kvm->stat.remote_tlb_flush;
Takuya Yoshikawa5befdc32014-02-18 17:22:47 +0900191 kvm->tlbs_dirty = false;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300192}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530193EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300194
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500195void kvm_reload_remote_mmus(struct kvm *kvm)
196{
Rusty Russell49846892008-12-08 20:26:24 +1030197 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500198}
199
Marcelo Tosattid8281992012-11-27 23:29:01 -0200200void kvm_make_mclock_inprogress_request(struct kvm *kvm)
201{
202 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
203}
204
Yang Zhang3d81bc72013-04-11 19:25:13 +0800205void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800206{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800207 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800208}
209
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000210int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
211{
212 struct page *page;
213 int r;
214
215 mutex_init(&vcpu->mutex);
216 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000217 vcpu->kvm = kvm;
218 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500219 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300220 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200221 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000222
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
224 if (!page) {
225 r = -ENOMEM;
226 goto fail;
227 }
228 vcpu->run = page_address(page);
229
Raghavendra K T4c088492012-07-18 19:07:46 +0530230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530232 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530233
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800234 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000235 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800236 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000237 return 0;
238
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000239fail_free_run:
240 free_page((unsigned long)vcpu->run);
241fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000242 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000243}
244EXPORT_SYMBOL_GPL(kvm_vcpu_init);
245
246void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
247{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500248 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800249 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000250 free_page((unsigned long)vcpu->run);
251}
252EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
253
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200254#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
256{
257 return container_of(mn, struct kvm, mmu_notifier);
258}
259
260static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
263{
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200265 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200266
267 /*
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
275 * before returning.
276 *
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
279 *
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
284 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200285 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200286 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900287
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200288 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200290 /* we've to flush the tlb before the pages can be freed */
291 if (need_tlb_flush)
292 kvm_flush_remote_tlbs(kvm);
293
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900294 spin_unlock(&kvm->mmu_lock);
295 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200296}
297
Izik Eidus3da0dd42009-09-23 21:47:18 +0300298static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
299 struct mm_struct *mm,
300 unsigned long address,
301 pte_t pte)
302{
303 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200304 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300305
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200306 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300307 spin_lock(&kvm->mmu_lock);
308 kvm->mmu_notifier_seq++;
309 kvm_set_spte_hva(kvm, address, pte);
310 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200311 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300312}
313
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200314static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
315 struct mm_struct *mm,
316 unsigned long start,
317 unsigned long end)
318{
319 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200320 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200321
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200322 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200323 spin_lock(&kvm->mmu_lock);
324 /*
325 * The count increase must become visible at unlock time as no
326 * spte can be established without taking the mmu_lock and
327 * count is also read inside the mmu_lock critical section.
328 */
329 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900330 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800331 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200332 /* we've to flush the tlb before the pages can be freed */
333 if (need_tlb_flush)
334 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900335
336 spin_unlock(&kvm->mmu_lock);
337 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200338}
339
340static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341 struct mm_struct *mm,
342 unsigned long start,
343 unsigned long end)
344{
345 struct kvm *kvm = mmu_notifier_to_kvm(mn);
346
347 spin_lock(&kvm->mmu_lock);
348 /*
349 * This sequence increase will notify the kvm page fault that
350 * the page that is going to be mapped in the spte could have
351 * been freed.
352 */
353 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000354 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200355 /*
356 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000357 * below count decrease, which is ensured by the smp_wmb above
358 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200359 */
360 kvm->mmu_notifier_count--;
361 spin_unlock(&kvm->mmu_lock);
362
363 BUG_ON(kvm->mmu_notifier_count < 0);
364}
365
366static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367 struct mm_struct *mm,
368 unsigned long address)
369{
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200371 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200372
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200373 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200374 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200375
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900376 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200377 if (young)
378 kvm_flush_remote_tlbs(kvm);
379
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900380 spin_unlock(&kvm->mmu_lock);
381 srcu_read_unlock(&kvm->srcu, idx);
382
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200383 return young;
384}
385
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800386static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
387 struct mm_struct *mm,
388 unsigned long address)
389{
390 struct kvm *kvm = mmu_notifier_to_kvm(mn);
391 int young, idx;
392
393 idx = srcu_read_lock(&kvm->srcu);
394 spin_lock(&kvm->mmu_lock);
395 young = kvm_test_age_hva(kvm, address);
396 spin_unlock(&kvm->mmu_lock);
397 srcu_read_unlock(&kvm->srcu, idx);
398
399 return young;
400}
401
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100402static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
403 struct mm_struct *mm)
404{
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800406 int idx;
407
408 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300409 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800410 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100411}
412
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200413static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
414 .invalidate_page = kvm_mmu_notifier_invalidate_page,
415 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
416 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
417 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800418 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300419 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100420 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200421};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200422
423static int kvm_init_mmu_notifier(struct kvm *kvm)
424{
425 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
426 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
427}
428
429#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
430
431static int kvm_init_mmu_notifier(struct kvm *kvm)
432{
433 return 0;
434}
435
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200436#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
437
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800438static void kvm_init_memslots_id(struct kvm *kvm)
439{
440 int i;
441 struct kvm_memslots *slots = kvm->memslots;
442
443 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800444 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800445}
446
Carsten Ottee08b9632012-01-04 10:25:20 +0100447static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800448{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100449 int r, i;
450 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800451
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100452 if (!kvm)
453 return ERR_PTR(-ENOMEM);
454
Carsten Ottee08b9632012-01-04 10:25:20 +0100455 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100456 if (r)
457 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200458
459 r = hardware_enable_all();
460 if (r)
461 goto out_err_nodisable;
462
Avi Kivity75858a82009-01-04 17:10:50 +0200463#ifdef CONFIG_HAVE_KVM_IRQCHIP
464 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300465 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200466#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800467
Alex Williamson1e702d92012-12-10 10:33:32 -0700468 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
469
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200470 r = -ENOMEM;
471 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
472 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100473 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800474 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200475 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100476 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200477 for (i = 0; i < KVM_NR_BUSES; i++) {
478 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
479 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100480 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200481 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200482 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700483
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900484 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200485 kvm->mm = current->mm;
486 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400487 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800488 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300489 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200490 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300491 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000492 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700493
494 r = kvm_init_mmu_notifier(kvm);
495 if (r)
496 goto out_err;
497
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200498 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000499 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200500 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100501
Avi Kivityf17abe92007-02-21 19:28:04 +0200502 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200503
504out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100505 cleanup_srcu_struct(&kvm->srcu);
506out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200507 hardware_disable_all();
508out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200509 for (i = 0; i < KVM_NR_BUSES; i++)
510 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200511 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100512 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200513 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200514}
515
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900516/*
517 * Avoid using vmalloc for a small buffer.
518 * Should not be used when the size is statically known.
519 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900520void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900521{
522 if (size > PAGE_SIZE)
523 return vzalloc(size);
524 else
525 return kzalloc(size, GFP_KERNEL);
526}
527
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900528void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900529{
530 if (is_vmalloc_addr(addr))
531 vfree(addr);
532 else
533 kfree(addr);
534}
535
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900536static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
537{
538 if (!memslot->dirty_bitmap)
539 return;
540
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900541 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900542 memslot->dirty_bitmap = NULL;
543}
544
Avi Kivity6aa8b732006-12-10 02:21:36 -0800545/*
546 * Free any memory in @free but not in @dont.
547 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530548static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800549 struct kvm_memory_slot *dont)
550{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800551 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900552 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800553
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530554 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300555
Avi Kivity6aa8b732006-12-10 02:21:36 -0800556 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557}
558
Stephen Hemminger79408762013-12-29 12:12:29 -0800559static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800560{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200561 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800562 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800564 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530565 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200566
567 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800568}
569
Scott Wood07f0a7b2013-04-25 14:11:23 +0000570static void kvm_destroy_devices(struct kvm *kvm)
571{
572 struct list_head *node, *tmp;
573
574 list_for_each_safe(node, tmp, &kvm->devices) {
575 struct kvm_device *dev =
576 list_entry(node, struct kvm_device, vm_node);
577
578 list_del(node);
579 dev->ops->destroy(dev);
580 }
581}
582
Avi Kivityf17abe92007-02-21 19:28:04 +0200583static void kvm_destroy_vm(struct kvm *kvm)
584{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200585 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200586 struct mm_struct *mm = kvm->mm;
587
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800588 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200589 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800590 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200591 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200592 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200593 for (i = 0; i < KVM_NR_BUSES; i++)
594 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200595 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200596#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200598#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300599 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200600#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800601 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000602 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100603 kvm_free_physmem(kvm);
604 cleanup_srcu_struct(&kvm->srcu);
605 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200606 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200607 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200608}
609
Izik Eidusd39f13b2008-03-30 16:01:25 +0300610void kvm_get_kvm(struct kvm *kvm)
611{
612 atomic_inc(&kvm->users_count);
613}
614EXPORT_SYMBOL_GPL(kvm_get_kvm);
615
616void kvm_put_kvm(struct kvm *kvm)
617{
618 if (atomic_dec_and_test(&kvm->users_count))
619 kvm_destroy_vm(kvm);
620}
621EXPORT_SYMBOL_GPL(kvm_put_kvm);
622
623
Avi Kivityf17abe92007-02-21 19:28:04 +0200624static int kvm_vm_release(struct inode *inode, struct file *filp)
625{
626 struct kvm *kvm = filp->private_data;
627
Gregory Haskins721eecb2009-05-20 10:30:49 -0400628 kvm_irqfd_release(kvm);
629
Izik Eidusd39f13b2008-03-30 16:01:25 +0300630 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800631 return 0;
632}
633
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900634/*
635 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900636 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900637 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900638static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
639{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900640#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900641 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900642
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900643 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900644 if (!memslot->dirty_bitmap)
645 return -ENOMEM;
646
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900647#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900648 return 0;
649}
650
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800651static int cmp_memslot(const void *slot1, const void *slot2)
652{
653 struct kvm_memory_slot *s1, *s2;
654
655 s1 = (struct kvm_memory_slot *)slot1;
656 s2 = (struct kvm_memory_slot *)slot2;
657
658 if (s1->npages < s2->npages)
659 return 1;
660 if (s1->npages > s2->npages)
661 return -1;
662
663 return 0;
664}
665
666/*
667 * Sort the memslots base on its size, so the larger slots
668 * will get better fit.
669 */
670static void sort_memslots(struct kvm_memslots *slots)
671{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800672 int i;
673
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800674 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
675 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800676
677 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
678 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800679}
680
Stephen Hemminger79408762013-12-29 12:12:29 -0800681static void update_memslots(struct kvm_memslots *slots,
682 struct kvm_memory_slot *new,
683 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800684{
685 if (new) {
686 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800687 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800688 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800689
Xiao Guangrong28a37542011-11-24 19:04:35 +0800690 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800691 if (new->npages != npages)
692 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800693 }
694
Alex Williamson116c14c2012-12-21 08:20:16 -0700695 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800696}
697
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800698static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
699{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800700 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
701
702#ifdef KVM_CAP_READONLY_MEM
703 valid_flags |= KVM_MEM_READONLY;
704#endif
705
706 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800707 return -EINVAL;
708
709 return 0;
710}
711
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200712static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
713 struct kvm_memslots *slots, struct kvm_memory_slot *new)
714{
715 struct kvm_memslots *old_memslots = kvm->memslots;
716
717 update_memslots(slots, new, kvm->memslots->generation);
718 rcu_assign_pointer(kvm->memslots, slots);
719 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900720
721 kvm_arch_memslots_updated(kvm);
722
723 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200724}
725
Avi Kivity6aa8b732006-12-10 02:21:36 -0800726/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800727 * Allocate some memory and give it an address in the guest physical address
728 * space.
729 *
730 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800731 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500732 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800733 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800734int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900735 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800736{
Avi Kivity8234b222010-12-27 12:08:45 +0200737 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800738 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200739 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900740 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800741 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700742 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900743 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800744
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800745 r = check_memory_region_flags(mem);
746 if (r)
747 goto out;
748
Avi Kivity6aa8b732006-12-10 02:21:36 -0800749 r = -EINVAL;
750 /* General sanity checks */
751 if (mem->memory_size & (PAGE_SIZE - 1))
752 goto out;
753 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
754 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900755 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900756 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900757 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200758 !access_ok(VERIFY_WRITE,
759 (void __user *)(unsigned long)mem->userspace_addr,
760 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600761 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800762 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800763 goto out;
764 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
765 goto out;
766
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900767 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800768 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
769 npages = mem->memory_size >> PAGE_SHIFT;
770
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900771 r = -EINVAL;
772 if (npages > KVM_MEM_MAX_NR_PAGES)
773 goto out;
774
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775 if (!npages)
776 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
777
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900778 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779
Avi Kivitye36d96f2010-06-21 10:56:36 +0300780 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800781 new.base_gfn = base_gfn;
782 new.npages = npages;
783 new.flags = mem->flags;
784
Avi Kivity6aa8b732006-12-10 02:21:36 -0800785 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900786 if (npages) {
787 if (!old.npages)
788 change = KVM_MR_CREATE;
789 else { /* Modify an existing slot. */
790 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900791 (npages != old.npages) ||
792 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900793 goto out;
794
795 if (base_gfn != old.base_gfn)
796 change = KVM_MR_MOVE;
797 else if (new.flags != old.flags)
798 change = KVM_MR_FLAGS_ONLY;
799 else { /* Nothing to change. */
800 r = 0;
801 goto out;
802 }
803 }
804 } else if (old.npages) {
805 change = KVM_MR_DELETE;
806 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900807 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800808
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900809 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900810 /* Check for overlaps */
811 r = -EEXIST;
812 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900813 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
814 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900815 continue;
816 if (!((base_gfn + npages <= slot->base_gfn) ||
817 (base_gfn >= slot->base_gfn + slot->npages)))
818 goto out;
819 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800820 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800821
Avi Kivity6aa8b732006-12-10 02:21:36 -0800822 /* Free page dirty bitmap if unneeded */
823 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000824 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800825
826 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900827 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900828 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900829
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530830 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900831 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800832 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200833
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834 /* Allocate page dirty bitmap if needed */
835 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900836 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800837 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800838 }
839
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900840 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200841 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100842 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
843 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200844 if (!slots)
845 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800846 slot = id_to_memslot(slots, mem->slot);
847 slot->flags |= KVM_MEMSLOT_INVALID;
848
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200849 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200850
Alex Williamsone40f1932012-12-10 10:32:57 -0700851 /* slot was deleted or moved, clear iommu mapping */
852 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300853 /* From this point no new shadow pages pointing to a deleted,
854 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200855 *
856 * validation of sp->gfn happens in:
857 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
858 * - kvm_is_visible_gfn (mmu_check_roots)
859 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300860 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700861 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200862 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300863
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900864 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200865 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700866 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200867
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200868 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700869 /*
870 * We can re-use the old_memslots from above, the only difference
871 * from the currently installed memslots is the invalid flag. This
872 * will get overwritten by update_memslots anyway.
873 */
874 if (!slots) {
875 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
876 GFP_KERNEL);
877 if (!slots)
878 goto out_free;
879 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200880
881 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900882 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200883 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900884 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200885 }
886
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200887 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800888
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900889 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800890
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530891 kvm_free_physmem_slot(kvm, &old, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200892 kfree(old_memslots);
893
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200894 /*
895 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800896 * un-mapped and re-mapped if their base changes. Since base change
897 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800898 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800899 * slots (size changes, userspace addr changes and read-only flag
900 * changes) is disallowed above, so any other attribute changes getting
901 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200902 */
903 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800904 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800905 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800906 }
907
Avi Kivity6aa8b732006-12-10 02:21:36 -0800908 return 0;
909
910out_slots:
911 kfree(slots);
912out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530913 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800914out:
915 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200916}
917EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800918
919int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900920 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800921{
922 int r;
923
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200924 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900925 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200926 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800927 return r;
928}
Izik Eidus210c7c42007-10-24 23:52:57 +0200929EXPORT_SYMBOL_GPL(kvm_set_memory_region);
930
Stephen Hemminger79408762013-12-29 12:12:29 -0800931static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
932 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200933{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700934 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200935 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900936 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800937}
938
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800939int kvm_get_dirty_log(struct kvm *kvm,
940 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800941{
942 struct kvm_memory_slot *memslot;
943 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900944 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945 unsigned long any = 0;
946
Avi Kivity6aa8b732006-12-10 02:21:36 -0800947 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700948 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800949 goto out;
950
Xiao Guangrong28a37542011-11-24 19:04:35 +0800951 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800952 r = -ENOENT;
953 if (!memslot->dirty_bitmap)
954 goto out;
955
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900956 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957
Uri Lublincd1a4a92007-02-22 16:43:09 +0200958 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800959 any = memslot->dirty_bitmap[i];
960
961 r = -EFAULT;
962 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
963 goto out;
964
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800965 if (any)
966 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967
968 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800970 return r;
971}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530972EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800973
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900974bool kvm_largepages_enabled(void)
975{
976 return largepages_enabled;
977}
978
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300979void kvm_disable_largepages(void)
980{
981 largepages_enabled = false;
982}
983EXPORT_SYMBOL_GPL(kvm_disable_largepages);
984
Gleb Natapov49c77542010-10-18 15:22:23 +0200985struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
986{
987 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
988}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300989EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800990
Izik Eiduse0d62c72007-10-24 23:57:46 +0200991int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
992{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800993 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +0200994
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700995 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800996 memslot->flags & KVM_MEMSLOT_INVALID)
997 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200998
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800999 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001000}
1001EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1002
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001003unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1004{
1005 struct vm_area_struct *vma;
1006 unsigned long addr, size;
1007
1008 size = PAGE_SIZE;
1009
1010 addr = gfn_to_hva(kvm, gfn);
1011 if (kvm_is_error_hva(addr))
1012 return PAGE_SIZE;
1013
1014 down_read(&current->mm->mmap_sem);
1015 vma = find_vma(current->mm, addr);
1016 if (!vma)
1017 goto out;
1018
1019 size = vma_kernel_pagesize(vma);
1020
1021out:
1022 up_read(&current->mm->mmap_sem);
1023
1024 return size;
1025}
1026
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001027static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1028{
1029 return slot->flags & KVM_MEM_READONLY;
1030}
1031
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001032static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1033 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001034{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001035 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001036 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001037
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001038 if (memslot_is_readonly(slot) && write)
1039 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001040
1041 if (nr_pages)
1042 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1043
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001044 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001045}
Xiao Guangrong48987782010-08-22 19:11:43 +08001046
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001047static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1048 gfn_t *nr_pages)
1049{
1050 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1051}
1052
1053unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001054 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055{
1056 return gfn_to_hva_many(slot, gfn, NULL);
1057}
1058EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1059
Xiao Guangrong48987782010-08-22 19:11:43 +08001060unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1061{
Gleb Natapov49c77542010-10-18 15:22:23 +02001062 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001063}
Sheng Yang0d150292008-04-25 21:44:50 +08001064EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001065
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001066/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001067 * If writable is set to false, the hva returned by this function is only
1068 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001069 */
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001070unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001071{
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001072 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001073 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1074
1075 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001076 *writable = !memslot_is_readonly(slot);
1077
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001078 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001079}
1080
1081static int kvm_read_hva(void *data, void __user *hva, int len)
1082{
1083 return __copy_from_user(data, hva, len);
1084}
1085
1086static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1087{
1088 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001089}
1090
Geoff Levand39369f72013-04-05 19:20:30 +00001091static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001092 unsigned long start, int write, struct page **page)
1093{
1094 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1095
1096 if (write)
1097 flags |= FOLL_WRITE;
1098
1099 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1100}
1101
Huang Yingfafc3db2011-01-30 11:15:49 +08001102static inline int check_user_page_hwpoison(unsigned long addr)
1103{
1104 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1105
1106 rc = __get_user_pages(current, current->mm, addr, 1,
1107 flags, NULL, NULL, NULL);
1108 return rc == -EHWPOISON;
1109}
1110
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001111/*
1112 * The atomic path to get the writable pfn which will be stored in @pfn,
1113 * true indicates success, otherwise false is returned.
1114 */
1115static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1116 bool write_fault, bool *writable, pfn_t *pfn)
1117{
1118 struct page *page[1];
1119 int npages;
1120
1121 if (!(async || atomic))
1122 return false;
1123
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001124 /*
1125 * Fast pin a writable pfn only if it is a write fault request
1126 * or the caller allows to map a writable pfn for a read fault
1127 * request.
1128 */
1129 if (!(write_fault || writable))
1130 return false;
1131
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001132 npages = __get_user_pages_fast(addr, 1, 1, page);
1133 if (npages == 1) {
1134 *pfn = page_to_pfn(page[0]);
1135
1136 if (writable)
1137 *writable = true;
1138 return true;
1139 }
1140
1141 return false;
1142}
1143
1144/*
1145 * The slow path to get the pfn of the specified host virtual address,
1146 * 1 indicates success, -errno is returned if error is detected.
1147 */
1148static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1149 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001150{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001151 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001152 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001153
1154 might_sleep();
1155
1156 if (writable)
1157 *writable = write_fault;
1158
1159 if (async) {
1160 down_read(&current->mm->mmap_sem);
1161 npages = get_user_page_nowait(current, current->mm,
1162 addr, write_fault, page);
1163 up_read(&current->mm->mmap_sem);
1164 } else
1165 npages = get_user_pages_fast(addr, 1, write_fault,
1166 page);
1167 if (npages != 1)
1168 return npages;
1169
1170 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001171 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001172 struct page *wpage[1];
1173
1174 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1175 if (npages == 1) {
1176 *writable = true;
1177 put_page(page[0]);
1178 page[0] = wpage[0];
1179 }
1180
1181 npages = 1;
1182 }
1183 *pfn = page_to_pfn(page[0]);
1184 return npages;
1185}
1186
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001187static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1188{
1189 if (unlikely(!(vma->vm_flags & VM_READ)))
1190 return false;
1191
1192 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1193 return false;
1194
1195 return true;
1196}
1197
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001198/*
1199 * Pin guest page in memory and return its pfn.
1200 * @addr: host virtual address which maps memory to the guest
1201 * @atomic: whether this function can sleep
1202 * @async: whether this function need to wait IO complete if the
1203 * host page is not in the memory
1204 * @write_fault: whether we should get a writable host page
1205 * @writable: whether it allows to map a writable host page for !@write_fault
1206 *
1207 * The function will map a writable host page for these two cases:
1208 * 1): @write_fault = true
1209 * 2): @write_fault = false && @writable, @writable will tell the caller
1210 * whether the mapping is writable.
1211 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001212static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1213 bool write_fault, bool *writable)
1214{
1215 struct vm_area_struct *vma;
1216 pfn_t pfn = 0;
1217 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001218
Gleb Natapovaf585b92010-10-14 11:22:46 +02001219 /* we can do it either atomically or asynchronously, not both */
1220 BUG_ON(atomic && async);
1221
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001222 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1223 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001224
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001225 if (atomic)
1226 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001227
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001228 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1229 if (npages == 1)
1230 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001231
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001232 down_read(&current->mm->mmap_sem);
1233 if (npages == -EHWPOISON ||
1234 (!async && check_user_page_hwpoison(addr))) {
1235 pfn = KVM_PFN_ERR_HWPOISON;
1236 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001237 }
Izik Eidus539cb662007-11-11 22:05:04 +02001238
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001239 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001240
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001241 if (vma == NULL)
1242 pfn = KVM_PFN_ERR_FAULT;
1243 else if ((vma->vm_flags & VM_PFNMAP)) {
1244 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1245 vma->vm_pgoff;
1246 BUG_ON(!kvm_is_mmio_pfn(pfn));
1247 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001248 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001249 *async = true;
1250 pfn = KVM_PFN_ERR_FAULT;
1251 }
1252exit:
1253 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001254 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001255}
1256
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001257static pfn_t
1258__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1259 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001260{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001261 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1262
1263 if (addr == KVM_HVA_ERR_RO_BAD)
1264 return KVM_PFN_ERR_RO_FAULT;
1265
1266 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001267 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001268
1269 /* Do not map writable pfn in the readonly memslot. */
1270 if (writable && memslot_is_readonly(slot)) {
1271 *writable = false;
1272 writable = NULL;
1273 }
1274
1275 return hva_to_pfn(addr, atomic, async, write_fault,
1276 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001277}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001278
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001279static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1280 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001281{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001282 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001283
Gleb Natapovaf585b92010-10-14 11:22:46 +02001284 if (async)
1285 *async = false;
1286
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001287 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001288
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001289 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1290 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001291}
1292
1293pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1294{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001295 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001296}
1297EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1298
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001299pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1300 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001301{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001302 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001303}
1304EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1305
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001306pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1307{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001308 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001309}
Anthony Liguori35149e22008-04-02 14:46:56 -05001310EXPORT_SYMBOL_GPL(gfn_to_pfn);
1311
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001312pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1313 bool *writable)
1314{
1315 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1316}
1317EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1318
Xiao Guangrongd5661042012-07-17 21:56:16 +08001319pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001320{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001321 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001322}
1323
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001324pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1325{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001326 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001327}
1328EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1329
Xiao Guangrong48987782010-08-22 19:11:43 +08001330int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1331 int nr_pages)
1332{
1333 unsigned long addr;
1334 gfn_t entry;
1335
Gleb Natapov49c77542010-10-18 15:22:23 +02001336 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001337 if (kvm_is_error_hva(addr))
1338 return -1;
1339
1340 if (entry < nr_pages)
1341 return 0;
1342
1343 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1344}
1345EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1346
Xiao Guangronga2766322012-07-26 11:58:59 +08001347static struct page *kvm_pfn_to_page(pfn_t pfn)
1348{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001349 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001350 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001351
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001352 if (kvm_is_mmio_pfn(pfn)) {
1353 WARN_ON(1);
1354 return KVM_ERR_PTR_BAD_PAGE;
1355 }
1356
Xiao Guangronga2766322012-07-26 11:58:59 +08001357 return pfn_to_page(pfn);
1358}
1359
Anthony Liguori35149e22008-04-02 14:46:56 -05001360struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1361{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001362 pfn_t pfn;
1363
1364 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001365
Xiao Guangronga2766322012-07-26 11:58:59 +08001366 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001367}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001368
Avi Kivity954bbbc2007-03-30 14:02:32 +03001369EXPORT_SYMBOL_GPL(gfn_to_page);
1370
Izik Eidusb4231d62007-11-20 11:49:33 +02001371void kvm_release_page_clean(struct page *page)
1372{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001373 WARN_ON(is_error_page(page));
1374
Anthony Liguori35149e22008-04-02 14:46:56 -05001375 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001376}
1377EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1378
Anthony Liguori35149e22008-04-02 14:46:56 -05001379void kvm_release_pfn_clean(pfn_t pfn)
1380{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001381 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001382 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001383}
1384EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1385
Izik Eidusb4231d62007-11-20 11:49:33 +02001386void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001387{
Xiao Guangronga2766322012-07-26 11:58:59 +08001388 WARN_ON(is_error_page(page));
1389
Anthony Liguori35149e22008-04-02 14:46:56 -05001390 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001391}
Izik Eidusb4231d62007-11-20 11:49:33 +02001392EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001393
Stephen Hemminger79408762013-12-29 12:12:29 -08001394static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001395{
1396 kvm_set_pfn_dirty(pfn);
1397 kvm_release_pfn_clean(pfn);
1398}
Anthony Liguori35149e22008-04-02 14:46:56 -05001399
1400void kvm_set_pfn_dirty(pfn_t pfn)
1401{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001402 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001403 struct page *page = pfn_to_page(pfn);
1404 if (!PageReserved(page))
1405 SetPageDirty(page);
1406 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001407}
1408EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1409
1410void kvm_set_pfn_accessed(pfn_t pfn)
1411{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001412 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001413 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001414}
1415EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1416
1417void kvm_get_pfn(pfn_t pfn)
1418{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001419 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001420 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001421}
1422EXPORT_SYMBOL_GPL(kvm_get_pfn);
1423
Izik Eidus195aefd2007-10-01 22:14:18 +02001424static int next_segment(unsigned long len, int offset)
1425{
1426 if (len > PAGE_SIZE - offset)
1427 return PAGE_SIZE - offset;
1428 else
1429 return len;
1430}
1431
1432int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1433 int len)
1434{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001435 int r;
1436 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001437
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001438 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001439 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001440 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001441 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001442 if (r)
1443 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001444 return 0;
1445}
1446EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1447
1448int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1449{
1450 gfn_t gfn = gpa >> PAGE_SHIFT;
1451 int seg;
1452 int offset = offset_in_page(gpa);
1453 int ret;
1454
1455 while ((seg = next_segment(len, offset)) != 0) {
1456 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1457 if (ret < 0)
1458 return ret;
1459 offset = 0;
1460 len -= seg;
1461 data += seg;
1462 ++gfn;
1463 }
1464 return 0;
1465}
1466EXPORT_SYMBOL_GPL(kvm_read_guest);
1467
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001468int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1469 unsigned long len)
1470{
1471 int r;
1472 unsigned long addr;
1473 gfn_t gfn = gpa >> PAGE_SHIFT;
1474 int offset = offset_in_page(gpa);
1475
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001476 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001477 if (kvm_is_error_hva(addr))
1478 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001479 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001480 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001481 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001482 if (r)
1483 return -EFAULT;
1484 return 0;
1485}
1486EXPORT_SYMBOL(kvm_read_guest_atomic);
1487
Izik Eidus195aefd2007-10-01 22:14:18 +02001488int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1489 int offset, int len)
1490{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001491 int r;
1492 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001493
Izik Eiduse0506bc2007-11-11 22:10:22 +02001494 addr = gfn_to_hva(kvm, gfn);
1495 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001496 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001497 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001498 if (r)
1499 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001500 mark_page_dirty(kvm, gfn);
1501 return 0;
1502}
1503EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1504
1505int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1506 unsigned long len)
1507{
1508 gfn_t gfn = gpa >> PAGE_SHIFT;
1509 int seg;
1510 int offset = offset_in_page(gpa);
1511 int ret;
1512
1513 while ((seg = next_segment(len, offset)) != 0) {
1514 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1515 if (ret < 0)
1516 return ret;
1517 offset = 0;
1518 len -= seg;
1519 data += seg;
1520 ++gfn;
1521 }
1522 return 0;
1523}
1524
Gleb Natapov49c77542010-10-18 15:22:23 +02001525int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001526 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001527{
1528 struct kvm_memslots *slots = kvm_memslots(kvm);
1529 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001530 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1531 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1532 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1533 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001534
1535 ghc->gpa = gpa;
1536 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001537 ghc->len = len;
1538 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1539 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1540 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001541 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001542 } else {
1543 /*
1544 * If the requested region crosses two memslots, we still
1545 * verify that the entire region is valid here.
1546 */
1547 while (start_gfn <= end_gfn) {
1548 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1549 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1550 &nr_pages_avail);
1551 if (kvm_is_error_hva(ghc->hva))
1552 return -EFAULT;
1553 start_gfn += nr_pages_avail;
1554 }
1555 /* Use the slow path for cross page reads and writes. */
1556 ghc->memslot = NULL;
1557 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001558 return 0;
1559}
1560EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1561
1562int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1563 void *data, unsigned long len)
1564{
1565 struct kvm_memslots *slots = kvm_memslots(kvm);
1566 int r;
1567
Andrew Honig8f964522013-03-29 09:35:21 -07001568 BUG_ON(len > ghc->len);
1569
Gleb Natapov49c77542010-10-18 15:22:23 +02001570 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001571 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1572
1573 if (unlikely(!ghc->memslot))
1574 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001575
1576 if (kvm_is_error_hva(ghc->hva))
1577 return -EFAULT;
1578
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001579 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001580 if (r)
1581 return -EFAULT;
1582 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1583
1584 return 0;
1585}
1586EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1587
Gleb Natapove03b6442011-07-11 15:28:11 -04001588int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1589 void *data, unsigned long len)
1590{
1591 struct kvm_memslots *slots = kvm_memslots(kvm);
1592 int r;
1593
Andrew Honig8f964522013-03-29 09:35:21 -07001594 BUG_ON(len > ghc->len);
1595
Gleb Natapove03b6442011-07-11 15:28:11 -04001596 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001597 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1598
1599 if (unlikely(!ghc->memslot))
1600 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001601
1602 if (kvm_is_error_hva(ghc->hva))
1603 return -EFAULT;
1604
1605 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1606 if (r)
1607 return -EFAULT;
1608
1609 return 0;
1610}
1611EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1612
Izik Eidus195aefd2007-10-01 22:14:18 +02001613int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1614{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001615 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1616
1617 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001618}
1619EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1620
1621int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1622{
1623 gfn_t gfn = gpa >> PAGE_SHIFT;
1624 int seg;
1625 int offset = offset_in_page(gpa);
1626 int ret;
1627
1628 while ((seg = next_segment(len, offset)) != 0) {
1629 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1630 if (ret < 0)
1631 return ret;
1632 offset = 0;
1633 len -= seg;
1634 ++gfn;
1635 }
1636 return 0;
1637}
1638EXPORT_SYMBOL_GPL(kvm_clear_guest);
1639
Stephen Hemminger79408762013-12-29 12:12:29 -08001640static void mark_page_dirty_in_slot(struct kvm *kvm,
1641 struct kvm_memory_slot *memslot,
1642 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001643{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001644 if (memslot && memslot->dirty_bitmap) {
1645 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001646
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001647 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001648 }
1649}
1650
Gleb Natapov49c77542010-10-18 15:22:23 +02001651void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1652{
1653 struct kvm_memory_slot *memslot;
1654
1655 memslot = gfn_to_memslot(kvm, gfn);
1656 mark_page_dirty_in_slot(kvm, memslot, gfn);
1657}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301658EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001659
Eddie Dongb6958ce2007-07-18 12:15:21 +03001660/*
1661 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001663void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001664{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001665 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001666
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001667 for (;;) {
1668 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001669
Gleb Natapova1b37102009-07-09 15:33:52 +03001670 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001671 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001672 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001673 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001674 if (kvm_cpu_has_pending_timer(vcpu))
1675 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001676 if (signal_pending(current))
1677 break;
1678
Eddie Dongb6958ce2007-07-18 12:15:21 +03001679 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001680 }
1681
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001682 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001683}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301684EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001685
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001686#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001687/*
1688 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1689 */
1690void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1691{
1692 int me;
1693 int cpu = vcpu->cpu;
1694 wait_queue_head_t *wqp;
1695
1696 wqp = kvm_arch_vcpu_wq(vcpu);
1697 if (waitqueue_active(wqp)) {
1698 wake_up_interruptible(wqp);
1699 ++vcpu->stat.halt_wakeup;
1700 }
1701
1702 me = get_cpu();
1703 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1704 if (kvm_arch_vcpu_should_kick(vcpu))
1705 smp_send_reschedule(cpu);
1706 put_cpu();
1707}
Yang Zhanga20ed542013-04-11 19:25:15 +08001708EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001709#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001710
Konstantin Weitz41628d32012-04-25 15:30:38 +02001711bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1712{
1713 struct pid *pid;
1714 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301715 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001716
1717 rcu_read_lock();
1718 pid = rcu_dereference(target->pid);
1719 if (pid)
1720 task = get_pid_task(target->pid, PIDTYPE_PID);
1721 rcu_read_unlock();
1722 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301723 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001724 if (task->flags & PF_VCPU) {
1725 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301726 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001727 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301728 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001729 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301730
1731 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001732}
1733EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1734
Raghavendra K T06e48c52012-07-19 15:17:52 +05301735/*
1736 * Helper that checks whether a VCPU is eligible for directed yield.
1737 * Most eligible candidate to yield is decided by following heuristics:
1738 *
1739 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1740 * (preempted lock holder), indicated by @in_spin_loop.
1741 * Set at the beiginning and cleared at the end of interception/PLE handler.
1742 *
1743 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1744 * chance last time (mostly it has become eligible now since we have probably
1745 * yielded to lockholder in last iteration. This is done by toggling
1746 * @dy_eligible each time a VCPU checked for eligibility.)
1747 *
1748 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1749 * to preempted lock-holder could result in wrong VCPU selection and CPU
1750 * burning. Giving priority for a potential lock-holder increases lock
1751 * progress.
1752 *
1753 * Since algorithm is based on heuristics, accessing another VCPU data without
1754 * locking does not harm. It may result in trying to yield to same VCPU, fail
1755 * and continue with next VCPU and so on.
1756 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001757static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301758{
Scott Wood4a55dd72014-01-09 18:43:16 -06001759#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301760 bool eligible;
1761
1762 eligible = !vcpu->spin_loop.in_spin_loop ||
1763 (vcpu->spin_loop.in_spin_loop &&
1764 vcpu->spin_loop.dy_eligible);
1765
1766 if (vcpu->spin_loop.in_spin_loop)
1767 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1768
1769 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001770#else
1771 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301772#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001773}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301774
Rik van Riel217ece62011-02-01 09:53:28 -05001775void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001776{
Rik van Riel217ece62011-02-01 09:53:28 -05001777 struct kvm *kvm = me->kvm;
1778 struct kvm_vcpu *vcpu;
1779 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1780 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301781 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001782 int pass;
1783 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001784
Raghavendra K T4c088492012-07-18 19:07:46 +05301785 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001786 /*
1787 * We boost the priority of a VCPU that is runnable but not
1788 * currently running, because it got preempted by something
1789 * else and called schedule in __vcpu_run. Hopefully that
1790 * VCPU is holding the lock that we need and will release it.
1791 * We approximate round-robin by starting at the last boosted VCPU.
1792 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301793 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001794 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001795 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001796 i = last_boosted_vcpu;
1797 continue;
1798 } else if (pass && i > last_boosted_vcpu)
1799 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301800 if (!ACCESS_ONCE(vcpu->preempted))
1801 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001802 if (vcpu == me)
1803 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001804 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001805 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301806 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1807 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301808
1809 yielded = kvm_vcpu_yield_to(vcpu);
1810 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001811 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001812 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301813 } else if (yielded < 0) {
1814 try--;
1815 if (!try)
1816 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001817 }
Rik van Riel217ece62011-02-01 09:53:28 -05001818 }
1819 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301820 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301821
1822 /* Ensure vcpu is not eligible during next spinloop */
1823 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001824}
1825EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1826
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001827static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001828{
1829 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001830 struct page *page;
1831
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001832 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001833 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001834#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001835 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001836 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001837#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001838#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1839 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1840 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1841#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001842 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001843 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001844 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001845 vmf->page = page;
1846 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001847}
1848
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001849static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001850 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001851};
1852
1853static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1854{
1855 vma->vm_ops = &kvm_vcpu_vm_ops;
1856 return 0;
1857}
1858
Avi Kivitybccf2152007-02-21 18:04:26 +02001859static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1860{
1861 struct kvm_vcpu *vcpu = filp->private_data;
1862
Al Viro66c0b392008-04-19 20:33:56 +01001863 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001864 return 0;
1865}
1866
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001867static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001868 .release = kvm_vcpu_release,
1869 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001870#ifdef CONFIG_COMPAT
1871 .compat_ioctl = kvm_vcpu_compat_ioctl,
1872#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001873 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001874 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001875};
1876
1877/*
1878 * Allocates an inode for the vcpu.
1879 */
1880static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1881{
Yann Droneaud24009b02013-08-24 22:14:07 +02001882 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001883}
1884
Avi Kivityc5ea7662007-02-20 18:41:05 +02001885/*
1886 * Creates some virtual cpus. Good luck creating more than one.
1887 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001888static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001889{
1890 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001891 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001892
Andy Honig338c7db2013-11-18 16:09:22 -08001893 if (id >= KVM_MAX_VCPUS)
1894 return -EINVAL;
1895
Gleb Natapov73880c82009-06-09 15:56:28 +03001896 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001897 if (IS_ERR(vcpu))
1898 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001899
Avi Kivity15ad7142007-07-11 18:17:21 +03001900 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1901
Avi Kivity26e52152007-11-20 15:30:24 +02001902 r = kvm_arch_vcpu_setup(vcpu);
1903 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001904 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001905
Shaohua Li11ec2802007-07-23 14:51:37 +08001906 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001907 if (!kvm_vcpu_compatible(vcpu)) {
1908 r = -EINVAL;
1909 goto unlock_vcpu_destroy;
1910 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001911 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1912 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001913 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001914 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001915
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001916 kvm_for_each_vcpu(r, v, kvm)
1917 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001918 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001919 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001920 }
1921
1922 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001923
1924 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001925 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001926 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001927 if (r < 0) {
1928 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001929 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001930 }
1931
1932 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1933 smp_wmb();
1934 atomic_inc(&kvm->online_vcpus);
1935
Gleb Natapov73880c82009-06-09 15:56:28 +03001936 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001937 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001938 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001939
Jan Kiszkad7805922011-05-23 10:33:05 +02001940unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001941 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001942vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001943 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001944 return r;
1945}
1946
Avi Kivity1961d272007-03-05 19:46:05 +02001947static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1948{
1949 if (sigset) {
1950 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1951 vcpu->sigset_active = 1;
1952 vcpu->sigset = *sigset;
1953 } else
1954 vcpu->sigset_active = 0;
1955 return 0;
1956}
1957
Avi Kivitybccf2152007-02-21 18:04:26 +02001958static long kvm_vcpu_ioctl(struct file *filp,
1959 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001960{
Avi Kivitybccf2152007-02-21 18:04:26 +02001961 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001962 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001963 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001964 struct kvm_fpu *fpu = NULL;
1965 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001966
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001967 if (vcpu->kvm->mm != current->mm)
1968 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001969
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001970#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001971 /*
1972 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1973 * so vcpu_load() would break it.
1974 */
1975 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1976 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1977#endif
1978
1979
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001980 r = vcpu_load(vcpu);
1981 if (r)
1982 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001983 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001984 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001985 r = -EINVAL;
1986 if (arg)
1987 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001988 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001989 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001990 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001991 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001992 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001993
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001994 r = -ENOMEM;
1995 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1996 if (!kvm_regs)
1997 goto out;
1998 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001999 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002000 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002001 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002002 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2003 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002004 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002005out_free1:
2006 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002007 break;
2008 }
2009 case KVM_SET_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;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002013 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2014 if (IS_ERR(kvm_regs)) {
2015 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002016 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002017 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002018 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002019 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002020 break;
2021 }
2022 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002023 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2024 r = -ENOMEM;
2025 if (!kvm_sregs)
2026 goto out;
2027 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002028 if (r)
2029 goto out;
2030 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002031 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002032 goto out;
2033 r = 0;
2034 break;
2035 }
2036 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002037 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2038 if (IS_ERR(kvm_sregs)) {
2039 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002040 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002041 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002042 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002043 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002044 break;
2045 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002046 case KVM_GET_MP_STATE: {
2047 struct kvm_mp_state mp_state;
2048
2049 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2050 if (r)
2051 goto out;
2052 r = -EFAULT;
2053 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2054 goto out;
2055 r = 0;
2056 break;
2057 }
2058 case KVM_SET_MP_STATE: {
2059 struct kvm_mp_state mp_state;
2060
2061 r = -EFAULT;
2062 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2063 goto out;
2064 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002065 break;
2066 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002067 case KVM_TRANSLATE: {
2068 struct kvm_translation tr;
2069
2070 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002071 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002072 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002073 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002074 if (r)
2075 goto out;
2076 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002077 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002078 goto out;
2079 r = 0;
2080 break;
2081 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002082 case KVM_SET_GUEST_DEBUG: {
2083 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002084
2085 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002086 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002087 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002088 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002089 break;
2090 }
Avi Kivity1961d272007-03-05 19:46:05 +02002091 case KVM_SET_SIGNAL_MASK: {
2092 struct kvm_signal_mask __user *sigmask_arg = argp;
2093 struct kvm_signal_mask kvm_sigmask;
2094 sigset_t sigset, *p;
2095
2096 p = NULL;
2097 if (argp) {
2098 r = -EFAULT;
2099 if (copy_from_user(&kvm_sigmask, argp,
2100 sizeof kvm_sigmask))
2101 goto out;
2102 r = -EINVAL;
2103 if (kvm_sigmask.len != sizeof sigset)
2104 goto out;
2105 r = -EFAULT;
2106 if (copy_from_user(&sigset, sigmask_arg->sigset,
2107 sizeof sigset))
2108 goto out;
2109 p = &sigset;
2110 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002111 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002112 break;
2113 }
Avi Kivityb8836732007-04-01 16:34:31 +03002114 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002115 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2116 r = -ENOMEM;
2117 if (!fpu)
2118 goto out;
2119 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002120 if (r)
2121 goto out;
2122 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002123 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002124 goto out;
2125 r = 0;
2126 break;
2127 }
2128 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002129 fpu = memdup_user(argp, sizeof(*fpu));
2130 if (IS_ERR(fpu)) {
2131 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002132 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002133 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002134 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002135 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002136 break;
2137 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002138 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002139 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002140 }
2141out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002142 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002143 kfree(fpu);
2144 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002145 return r;
2146}
2147
Alexander Graf1dda6062011-06-08 02:45:37 +02002148#ifdef CONFIG_COMPAT
2149static long kvm_vcpu_compat_ioctl(struct file *filp,
2150 unsigned int ioctl, unsigned long arg)
2151{
2152 struct kvm_vcpu *vcpu = filp->private_data;
2153 void __user *argp = compat_ptr(arg);
2154 int r;
2155
2156 if (vcpu->kvm->mm != current->mm)
2157 return -EIO;
2158
2159 switch (ioctl) {
2160 case KVM_SET_SIGNAL_MASK: {
2161 struct kvm_signal_mask __user *sigmask_arg = argp;
2162 struct kvm_signal_mask kvm_sigmask;
2163 compat_sigset_t csigset;
2164 sigset_t sigset;
2165
2166 if (argp) {
2167 r = -EFAULT;
2168 if (copy_from_user(&kvm_sigmask, argp,
2169 sizeof kvm_sigmask))
2170 goto out;
2171 r = -EINVAL;
2172 if (kvm_sigmask.len != sizeof csigset)
2173 goto out;
2174 r = -EFAULT;
2175 if (copy_from_user(&csigset, sigmask_arg->sigset,
2176 sizeof csigset))
2177 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002178 sigset_from_compat(&sigset, &csigset);
2179 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2180 } else
2181 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002182 break;
2183 }
2184 default:
2185 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2186 }
2187
2188out:
2189 return r;
2190}
2191#endif
2192
Scott Wood852b6d52013-04-12 14:08:42 +00002193static int kvm_device_ioctl_attr(struct kvm_device *dev,
2194 int (*accessor)(struct kvm_device *dev,
2195 struct kvm_device_attr *attr),
2196 unsigned long arg)
2197{
2198 struct kvm_device_attr attr;
2199
2200 if (!accessor)
2201 return -EPERM;
2202
2203 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2204 return -EFAULT;
2205
2206 return accessor(dev, &attr);
2207}
2208
2209static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2210 unsigned long arg)
2211{
2212 struct kvm_device *dev = filp->private_data;
2213
2214 switch (ioctl) {
2215 case KVM_SET_DEVICE_ATTR:
2216 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2217 case KVM_GET_DEVICE_ATTR:
2218 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2219 case KVM_HAS_DEVICE_ATTR:
2220 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2221 default:
2222 if (dev->ops->ioctl)
2223 return dev->ops->ioctl(dev, ioctl, arg);
2224
2225 return -ENOTTY;
2226 }
2227}
2228
Scott Wood852b6d52013-04-12 14:08:42 +00002229static int kvm_device_release(struct inode *inode, struct file *filp)
2230{
2231 struct kvm_device *dev = filp->private_data;
2232 struct kvm *kvm = dev->kvm;
2233
Scott Wood852b6d52013-04-12 14:08:42 +00002234 kvm_put_kvm(kvm);
2235 return 0;
2236}
2237
2238static const struct file_operations kvm_device_fops = {
2239 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002240#ifdef CONFIG_COMPAT
2241 .compat_ioctl = kvm_device_ioctl,
2242#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002243 .release = kvm_device_release,
2244};
2245
2246struct kvm_device *kvm_device_from_filp(struct file *filp)
2247{
2248 if (filp->f_op != &kvm_device_fops)
2249 return NULL;
2250
2251 return filp->private_data;
2252}
2253
2254static int kvm_ioctl_create_device(struct kvm *kvm,
2255 struct kvm_create_device *cd)
2256{
2257 struct kvm_device_ops *ops = NULL;
2258 struct kvm_device *dev;
2259 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2260 int ret;
2261
2262 switch (cd->type) {
Scott Wood5df554a2013-04-12 14:08:46 +00002263#ifdef CONFIG_KVM_MPIC
2264 case KVM_DEV_TYPE_FSL_MPIC_20:
2265 case KVM_DEV_TYPE_FSL_MPIC_42:
2266 ops = &kvm_mpic_ops;
2267 break;
2268#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002269#ifdef CONFIG_KVM_XICS
2270 case KVM_DEV_TYPE_XICS:
2271 ops = &kvm_xics_ops;
2272 break;
2273#endif
Alex Williamsonec535002013-10-30 11:02:17 -06002274#ifdef CONFIG_KVM_VFIO
2275 case KVM_DEV_TYPE_VFIO:
2276 ops = &kvm_vfio_ops;
2277 break;
2278#endif
Christoffer Dall73306722013-10-25 17:29:18 +01002279#ifdef CONFIG_KVM_ARM_VGIC
2280 case KVM_DEV_TYPE_ARM_VGIC_V2:
2281 ops = &kvm_arm_vgic_v2_ops;
2282 break;
2283#endif
Jens Freimannc05c4182013-10-07 16:13:45 +02002284#ifdef CONFIG_S390
2285 case KVM_DEV_TYPE_FLIC:
2286 ops = &kvm_flic_ops;
2287 break;
2288#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002289 default:
2290 return -ENODEV;
2291 }
2292
2293 if (test)
2294 return 0;
2295
2296 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2297 if (!dev)
2298 return -ENOMEM;
2299
2300 dev->ops = ops;
2301 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002302
2303 ret = ops->create(dev, cd->type);
2304 if (ret < 0) {
2305 kfree(dev);
2306 return ret;
2307 }
2308
Yann Droneaud24009b02013-08-24 22:14:07 +02002309 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002310 if (ret < 0) {
2311 ops->destroy(dev);
2312 return ret;
2313 }
2314
Scott Wood07f0a7b2013-04-25 14:11:23 +00002315 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002316 kvm_get_kvm(kvm);
2317 cd->fd = ret;
2318 return 0;
2319}
2320
Avi Kivitybccf2152007-02-21 18:04:26 +02002321static long kvm_vm_ioctl(struct file *filp,
2322 unsigned int ioctl, unsigned long arg)
2323{
2324 struct kvm *kvm = filp->private_data;
2325 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002326 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002327
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002328 if (kvm->mm != current->mm)
2329 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002330 switch (ioctl) {
2331 case KVM_CREATE_VCPU:
2332 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002333 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002334 case KVM_SET_USER_MEMORY_REGION: {
2335 struct kvm_userspace_memory_region kvm_userspace_mem;
2336
2337 r = -EFAULT;
2338 if (copy_from_user(&kvm_userspace_mem, argp,
2339 sizeof kvm_userspace_mem))
2340 goto out;
2341
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002342 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002343 break;
2344 }
2345 case KVM_GET_DIRTY_LOG: {
2346 struct kvm_dirty_log log;
2347
2348 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002349 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002350 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002351 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002352 break;
2353 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002354#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2355 case KVM_REGISTER_COALESCED_MMIO: {
2356 struct kvm_coalesced_mmio_zone zone;
2357 r = -EFAULT;
2358 if (copy_from_user(&zone, argp, sizeof zone))
2359 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002360 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002361 break;
2362 }
2363 case KVM_UNREGISTER_COALESCED_MMIO: {
2364 struct kvm_coalesced_mmio_zone zone;
2365 r = -EFAULT;
2366 if (copy_from_user(&zone, argp, sizeof zone))
2367 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002368 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002369 break;
2370 }
2371#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002372 case KVM_IRQFD: {
2373 struct kvm_irqfd data;
2374
2375 r = -EFAULT;
2376 if (copy_from_user(&data, argp, sizeof data))
2377 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002378 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002379 break;
2380 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002381 case KVM_IOEVENTFD: {
2382 struct kvm_ioeventfd data;
2383
2384 r = -EFAULT;
2385 if (copy_from_user(&data, argp, sizeof data))
2386 goto out;
2387 r = kvm_ioeventfd(kvm, &data);
2388 break;
2389 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002390#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2391 case KVM_SET_BOOT_CPU_ID:
2392 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002393 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002394 if (atomic_read(&kvm->online_vcpus) != 0)
2395 r = -EBUSY;
2396 else
2397 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002398 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002399 break;
2400#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002401#ifdef CONFIG_HAVE_KVM_MSI
2402 case KVM_SIGNAL_MSI: {
2403 struct kvm_msi msi;
2404
2405 r = -EFAULT;
2406 if (copy_from_user(&msi, argp, sizeof msi))
2407 goto out;
2408 r = kvm_send_userspace_msi(kvm, &msi);
2409 break;
2410 }
2411#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002412#ifdef __KVM_HAVE_IRQ_LINE
2413 case KVM_IRQ_LINE_STATUS:
2414 case KVM_IRQ_LINE: {
2415 struct kvm_irq_level irq_event;
2416
2417 r = -EFAULT;
2418 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2419 goto out;
2420
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002421 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2422 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002423 if (r)
2424 goto out;
2425
2426 r = -EFAULT;
2427 if (ioctl == KVM_IRQ_LINE_STATUS) {
2428 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2429 goto out;
2430 }
2431
2432 r = 0;
2433 break;
2434 }
2435#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002436#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2437 case KVM_SET_GSI_ROUTING: {
2438 struct kvm_irq_routing routing;
2439 struct kvm_irq_routing __user *urouting;
2440 struct kvm_irq_routing_entry *entries;
2441
2442 r = -EFAULT;
2443 if (copy_from_user(&routing, argp, sizeof(routing)))
2444 goto out;
2445 r = -EINVAL;
2446 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2447 goto out;
2448 if (routing.flags)
2449 goto out;
2450 r = -ENOMEM;
2451 entries = vmalloc(routing.nr * sizeof(*entries));
2452 if (!entries)
2453 goto out;
2454 r = -EFAULT;
2455 urouting = argp;
2456 if (copy_from_user(entries, urouting->entries,
2457 routing.nr * sizeof(*entries)))
2458 goto out_free_irq_routing;
2459 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2460 routing.flags);
2461 out_free_irq_routing:
2462 vfree(entries);
2463 break;
2464 }
2465#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002466 case KVM_CREATE_DEVICE: {
2467 struct kvm_create_device cd;
2468
2469 r = -EFAULT;
2470 if (copy_from_user(&cd, argp, sizeof(cd)))
2471 goto out;
2472
2473 r = kvm_ioctl_create_device(kvm, &cd);
2474 if (r)
2475 goto out;
2476
2477 r = -EFAULT;
2478 if (copy_to_user(argp, &cd, sizeof(cd)))
2479 goto out;
2480
2481 r = 0;
2482 break;
2483 }
Avi Kivityf17abe92007-02-21 19:28:04 +02002484 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002485 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002486 if (r == -ENOTTY)
2487 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002488 }
2489out:
2490 return r;
2491}
2492
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002493#ifdef CONFIG_COMPAT
2494struct compat_kvm_dirty_log {
2495 __u32 slot;
2496 __u32 padding1;
2497 union {
2498 compat_uptr_t dirty_bitmap; /* one bit per page */
2499 __u64 padding2;
2500 };
2501};
2502
2503static long kvm_vm_compat_ioctl(struct file *filp,
2504 unsigned int ioctl, unsigned long arg)
2505{
2506 struct kvm *kvm = filp->private_data;
2507 int r;
2508
2509 if (kvm->mm != current->mm)
2510 return -EIO;
2511 switch (ioctl) {
2512 case KVM_GET_DIRTY_LOG: {
2513 struct compat_kvm_dirty_log compat_log;
2514 struct kvm_dirty_log log;
2515
2516 r = -EFAULT;
2517 if (copy_from_user(&compat_log, (void __user *)arg,
2518 sizeof(compat_log)))
2519 goto out;
2520 log.slot = compat_log.slot;
2521 log.padding1 = compat_log.padding1;
2522 log.padding2 = compat_log.padding2;
2523 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2524
2525 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002526 break;
2527 }
2528 default:
2529 r = kvm_vm_ioctl(filp, ioctl, arg);
2530 }
2531
2532out:
2533 return r;
2534}
2535#endif
2536
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002537static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002538 .release = kvm_vm_release,
2539 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002540#ifdef CONFIG_COMPAT
2541 .compat_ioctl = kvm_vm_compat_ioctl,
2542#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002543 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002544};
2545
Carsten Ottee08b9632012-01-04 10:25:20 +01002546static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002547{
Heiko Carstensaac87632010-10-27 17:22:10 +02002548 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002549 struct kvm *kvm;
2550
Carsten Ottee08b9632012-01-04 10:25:20 +01002551 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002552 if (IS_ERR(kvm))
2553 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002554#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2555 r = kvm_coalesced_mmio_init(kvm);
2556 if (r < 0) {
2557 kvm_put_kvm(kvm);
2558 return r;
2559 }
2560#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002561 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002562 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002563 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002564
Heiko Carstensaac87632010-10-27 17:22:10 +02002565 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002566}
2567
Avi Kivity1a811b62008-12-08 18:25:27 +02002568static long kvm_dev_ioctl_check_extension_generic(long arg)
2569{
2570 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002571 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002572 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002573 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002574#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2575 case KVM_CAP_SET_BOOT_CPU_ID:
2576#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002577 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002578#ifdef CONFIG_HAVE_KVM_MSI
2579 case KVM_CAP_SIGNAL_MSI:
2580#endif
Alexander Graf7df35f52013-04-16 12:12:49 +02002581#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2582 case KVM_CAP_IRQFD_RESAMPLE:
2583#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002584 return 1;
Alexander Grafa725d562013-04-17 13:29:30 +02002585#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002586 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002587 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002588#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002589 default:
2590 break;
2591 }
2592 return kvm_dev_ioctl_check_extension(arg);
2593}
2594
Avi Kivityf17abe92007-02-21 19:28:04 +02002595static long kvm_dev_ioctl(struct file *filp,
2596 unsigned int ioctl, unsigned long arg)
2597{
Avi Kivity07c45a32007-03-07 13:05:38 +02002598 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002599
2600 switch (ioctl) {
2601 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002602 r = -EINVAL;
2603 if (arg)
2604 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002605 r = KVM_API_VERSION;
2606 break;
2607 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002608 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002609 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002610 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002611 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002612 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002613 case KVM_GET_VCPU_MMAP_SIZE:
2614 r = -EINVAL;
2615 if (arg)
2616 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002617 r = PAGE_SIZE; /* struct kvm_run */
2618#ifdef CONFIG_X86
2619 r += PAGE_SIZE; /* pio data page */
2620#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002621#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2622 r += PAGE_SIZE; /* coalesced mmio ring page */
2623#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002624 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002625 case KVM_TRACE_ENABLE:
2626 case KVM_TRACE_PAUSE:
2627 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002628 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002629 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002630 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002631 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002632 }
2633out:
2634 return r;
2635}
2636
Avi Kivity6aa8b732006-12-10 02:21:36 -08002637static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002638 .unlocked_ioctl = kvm_dev_ioctl,
2639 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002640 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002641};
2642
2643static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002644 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002645 "kvm",
2646 &kvm_chardev_ops,
2647};
2648
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002649static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002650{
2651 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002652 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002653
Rusty Russell7f59f492008-12-07 21:25:45 +10302654 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002655 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002656
Rusty Russell7f59f492008-12-07 21:25:45 +10302657 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002658
2659 r = kvm_arch_hardware_enable(NULL);
2660
2661 if (r) {
2662 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2663 atomic_inc(&hardware_enable_failed);
2664 printk(KERN_INFO "kvm: enabling virtualization on "
2665 "CPU%d failed\n", cpu);
2666 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002667}
2668
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002669static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002670{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002671 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002672 if (kvm_usage_count)
2673 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002674 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002675}
2676
2677static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002678{
2679 int cpu = raw_smp_processor_id();
2680
Rusty Russell7f59f492008-12-07 21:25:45 +10302681 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002682 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302683 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002684 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002685}
2686
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002687static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002688{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002689 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002690 if (kvm_usage_count)
2691 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002692 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002693}
2694
Alexander Graf10474ae2009-09-15 11:37:46 +02002695static void hardware_disable_all_nolock(void)
2696{
2697 BUG_ON(!kvm_usage_count);
2698
2699 kvm_usage_count--;
2700 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002701 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002702}
2703
2704static void hardware_disable_all(void)
2705{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002706 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002707 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002708 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002709}
2710
2711static int hardware_enable_all(void)
2712{
2713 int r = 0;
2714
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002715 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002716
2717 kvm_usage_count++;
2718 if (kvm_usage_count == 1) {
2719 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002720 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002721
2722 if (atomic_read(&hardware_enable_failed)) {
2723 hardware_disable_all_nolock();
2724 r = -EBUSY;
2725 }
2726 }
2727
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002728 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002729
2730 return r;
2731}
2732
Avi Kivity774c47f2007-02-12 00:54:47 -08002733static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2734 void *v)
2735{
2736 int cpu = (long)v;
2737
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002738 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002739 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002740 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002741 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2742 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002743 hardware_disable();
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002744 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002745 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002746 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2747 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002748 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002749 break;
2750 }
2751 return NOTIFY_OK;
2752}
2753
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002754static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002755 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002756{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002757 /*
2758 * Some (well, at least mine) BIOSes hang on reboot if
2759 * in vmx root mode.
2760 *
2761 * And Intel TXT required VMX off for all cpu when system shutdown.
2762 */
2763 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2764 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002765 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002766 return NOTIFY_OK;
2767}
2768
2769static struct notifier_block kvm_reboot_notifier = {
2770 .notifier_call = kvm_reboot,
2771 .priority = 0,
2772};
2773
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002774static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002775{
2776 int i;
2777
2778 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002779 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002780
2781 kvm_iodevice_destructor(pos);
2782 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002783 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002784}
2785
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002786static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2787 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002788{
Sasha Levin743eeb02011-07-27 16:00:48 +03002789 if (r1->addr < r2->addr)
2790 return -1;
2791 if (r1->addr + r1->len > r2->addr + r2->len)
2792 return 1;
2793 return 0;
2794}
2795
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002796static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2797{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002798 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002799}
2800
Geoff Levand39369f72013-04-05 19:20:30 +00002801static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002802 gpa_t addr, int len)
2803{
Sasha Levin743eeb02011-07-27 16:00:48 +03002804 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2805 .addr = addr,
2806 .len = len,
2807 .dev = dev,
2808 };
2809
2810 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2811 kvm_io_bus_sort_cmp, NULL);
2812
2813 return 0;
2814}
2815
Geoff Levand39369f72013-04-05 19:20:30 +00002816static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002817 gpa_t addr, int len)
2818{
2819 struct kvm_io_range *range, key;
2820 int off;
2821
2822 key = (struct kvm_io_range) {
2823 .addr = addr,
2824 .len = len,
2825 };
2826
2827 range = bsearch(&key, bus->range, bus->dev_count,
2828 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2829 if (range == NULL)
2830 return -ENOENT;
2831
2832 off = range - bus->range;
2833
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002834 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002835 off--;
2836
2837 return off;
2838}
2839
Cornelia Huck126a5af2013-07-03 16:30:53 +02002840static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2841 struct kvm_io_range *range, const void *val)
2842{
2843 int idx;
2844
2845 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2846 if (idx < 0)
2847 return -EOPNOTSUPP;
2848
2849 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002850 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002851 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2852 range->len, val))
2853 return idx;
2854 idx++;
2855 }
2856
2857 return -EOPNOTSUPP;
2858}
2859
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002860/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002861int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002862 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002863{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002864 struct kvm_io_bus *bus;
2865 struct kvm_io_range range;
2866 int r;
2867
2868 range = (struct kvm_io_range) {
2869 .addr = addr,
2870 .len = len,
2871 };
2872
2873 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2874 r = __kvm_io_bus_write(bus, &range, val);
2875 return r < 0 ? r : 0;
2876}
2877
2878/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2879int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2880 int len, const void *val, long cookie)
2881{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002882 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002883 struct kvm_io_range range;
2884
2885 range = (struct kvm_io_range) {
2886 .addr = addr,
2887 .len = len,
2888 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002889
2890 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002891
2892 /* First try the device referenced by cookie. */
2893 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002894 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002895 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2896 val))
2897 return cookie;
2898
2899 /*
2900 * cookie contained garbage; fall back to search and return the
2901 * correct cookie value.
2902 */
2903 return __kvm_io_bus_write(bus, &range, val);
2904}
2905
2906static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2907 void *val)
2908{
2909 int idx;
2910
2911 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002912 if (idx < 0)
2913 return -EOPNOTSUPP;
2914
2915 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002916 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002917 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2918 range->len, val))
2919 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03002920 idx++;
2921 }
2922
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002923 return -EOPNOTSUPP;
2924}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002925
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002926/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002927int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2928 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002929{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002930 struct kvm_io_bus *bus;
2931 struct kvm_io_range range;
2932 int r;
2933
2934 range = (struct kvm_io_range) {
2935 .addr = addr,
2936 .len = len,
2937 };
2938
2939 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2940 r = __kvm_io_bus_read(bus, &range, val);
2941 return r < 0 ? r : 0;
2942}
2943
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002944
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002945/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002946int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2947 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002948{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002949 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002950
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002951 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08002952 /* exclude ioeventfd which is limited by maximum fd */
2953 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002954 return -ENOSPC;
2955
Amos Konga13007162012-03-09 12:17:32 +08002956 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2957 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002958 if (!new_bus)
2959 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002960 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2961 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002962 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002963 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2964 synchronize_srcu_expedited(&kvm->srcu);
2965 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002966
2967 return 0;
2968}
2969
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002970/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002971int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2972 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002973{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002974 int i, r;
2975 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002976
Sasha Levincdfca7b2011-12-04 19:36:28 +02002977 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002978 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002979 for (i = 0; i < bus->dev_count; i++)
2980 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002981 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002982 break;
2983 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002984
Amos Konga13007162012-03-09 12:17:32 +08002985 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002986 return r;
Amos Konga13007162012-03-09 12:17:32 +08002987
2988 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2989 sizeof(struct kvm_io_range)), GFP_KERNEL);
2990 if (!new_bus)
2991 return -ENOMEM;
2992
2993 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2994 new_bus->dev_count--;
2995 memcpy(new_bus->range + i, bus->range + i + 1,
2996 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002997
2998 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2999 synchronize_srcu_expedited(&kvm->srcu);
3000 kfree(bus);
3001 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003002}
3003
Avi Kivity774c47f2007-02-12 00:54:47 -08003004static struct notifier_block kvm_cpu_notifier = {
3005 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003006};
3007
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003008static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003009{
3010 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003011 struct kvm *kvm;
3012
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003013 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003014 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003015 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003016 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003017 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003018 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003019}
3020
3021DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3022
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003023static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003024{
3025 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003026 struct kvm *kvm;
3027 struct kvm_vcpu *vcpu;
3028 int i;
3029
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003030 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003031 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003032 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003033 kvm_for_each_vcpu(i, vcpu, kvm)
3034 *val += *(u32 *)((void *)vcpu + offset);
3035
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003036 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003037 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003038}
3039
Avi Kivityba1389b2007-11-18 16:24:12 +02003040DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3041
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003042static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003043 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3044 [KVM_STAT_VM] = &vm_stat_fops,
3045};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003046
Hamo4f69b682011-12-15 14:23:16 +08003047static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003048{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003049 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003050 struct kvm_stats_debugfs_item *p;
3051
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003052 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003053 if (kvm_debugfs_dir == NULL)
3054 goto out;
3055
3056 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003057 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003058 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003059 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003060 if (p->dentry == NULL)
3061 goto out_dir;
3062 }
3063
3064 return 0;
3065
3066out_dir:
3067 debugfs_remove_recursive(kvm_debugfs_dir);
3068out:
3069 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003070}
3071
3072static void kvm_exit_debug(void)
3073{
3074 struct kvm_stats_debugfs_item *p;
3075
3076 for (p = debugfs_entries; p->name; ++p)
3077 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003078 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003079}
3080
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003081static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003082{
Alexander Graf10474ae2009-09-15 11:37:46 +02003083 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003084 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003085 return 0;
3086}
3087
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003088static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003089{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003090 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003091 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003092 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003093 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003094}
3095
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003096static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003097 .suspend = kvm_suspend,
3098 .resume = kvm_resume,
3099};
3100
Avi Kivity15ad7142007-07-11 18:17:21 +03003101static inline
3102struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3103{
3104 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3105}
3106
3107static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3108{
3109 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303110 if (vcpu->preempted)
3111 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003112
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003113 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003114}
3115
3116static void kvm_sched_out(struct preempt_notifier *pn,
3117 struct task_struct *next)
3118{
3119 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3120
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303121 if (current->state == TASK_RUNNING)
3122 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003123 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003124}
3125
Avi Kivity0ee75be2010-04-28 15:39:01 +03003126int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003127 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003128{
3129 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003130 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003131
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003132 r = kvm_arch_init(opaque);
3133 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003134 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003135
Asias He7dac16c2013-05-08 10:57:29 +08003136 /*
3137 * kvm_arch_init makes sure there's at most one caller
3138 * for architectures that support multiple implementations,
3139 * like intel and amd on x86.
3140 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3141 * conflicts in case kvm is already setup for another implementation.
3142 */
3143 r = kvm_irqfd_init();
3144 if (r)
3145 goto out_irqfd;
3146
Avi Kivity8437a612009-06-06 14:52:35 -07003147 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303148 r = -ENOMEM;
3149 goto out_free_0;
3150 }
3151
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003152 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003153 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303154 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003155
Yang, Sheng002c7f72007-07-31 14:23:01 +03003156 for_each_online_cpu(cpu) {
3157 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003158 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003159 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003160 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003161 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003162 }
3163
Avi Kivity774c47f2007-02-12 00:54:47 -08003164 r = register_cpu_notifier(&kvm_cpu_notifier);
3165 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003166 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003167 register_reboot_notifier(&kvm_reboot_notifier);
3168
Rusty Russellc16f8622007-07-30 21:12:19 +10003169 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003170 if (!vcpu_align)
3171 vcpu_align = __alignof__(struct kvm_vcpu);
3172 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003173 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003174 if (!kvm_vcpu_cache) {
3175 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003176 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003177 }
3178
Gleb Natapovaf585b92010-10-14 11:22:46 +02003179 r = kvm_async_pf_init();
3180 if (r)
3181 goto out_free;
3182
Avi Kivity6aa8b732006-12-10 02:21:36 -08003183 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003184 kvm_vm_fops.owner = module;
3185 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003186
3187 r = misc_register(&kvm_dev);
3188 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003189 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003190 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003191 }
3192
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003193 register_syscore_ops(&kvm_syscore_ops);
3194
Avi Kivity15ad7142007-07-11 18:17:21 +03003195 kvm_preempt_ops.sched_in = kvm_sched_in;
3196 kvm_preempt_ops.sched_out = kvm_sched_out;
3197
Hamo4f69b682011-12-15 14:23:16 +08003198 r = kvm_init_debug();
3199 if (r) {
3200 printk(KERN_ERR "kvm: create debugfs files failed\n");
3201 goto out_undebugfs;
3202 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003203
Avi Kivityc7addb92007-09-16 18:58:32 +02003204 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003205
Hamo4f69b682011-12-15 14:23:16 +08003206out_undebugfs:
3207 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003208 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003209out_unreg:
3210 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003211out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003212 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003213out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003214 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003215 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003216out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003217out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003218 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303219out_free_0a:
3220 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003221out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003222 kvm_irqfd_exit();
3223out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003224 kvm_arch_exit();
3225out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003226 return r;
3227}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003228EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003229
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003230void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003231{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003232 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003233 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003234 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003235 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003236 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003237 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003238 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003239 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003240 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003241 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003242 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303243 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003244}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003245EXPORT_SYMBOL_GPL(kvm_exit);