blob: f5283438ee05e165b50b693c2d864f248d82a90d [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
David Matlack2ea75be2014-09-19 16:03:25 -070055#include <asm/ioctl.h>
Avi Kivitye4956062007-06-28 14:15:57 -040056#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020057#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080058
Laurent Vivier5f94c172008-05-30 16:05:54 +020059#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020060#include "async_pf.h"
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +020061#include "vfio.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020062
Marcelo Tosatti229456f2009-06-17 09:22:14 -030063#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h>
65
Avi Kivity6aa8b732006-12-10 02:21:36 -080066MODULE_AUTHOR("Qumranet");
67MODULE_LICENSE("GPL");
68
Marcelo Tosattifa40a822009-06-04 15:08:24 -030069/*
70 * Ordering of locks:
71 *
Sheng Yangfae3a352009-12-15 10:28:07 +080072 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030073 */
74
Paolo Bonzini2f303b72013-09-25 13:53:07 +020075DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020076static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080077LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080078
Rusty Russell7f59f492008-12-07 21:25:45 +103079static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020080static int kvm_usage_count = 0;
81static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030082
Rusty Russellc16f8622007-07-30 21:12:19 +100083struct kmem_cache *kvm_vcpu_cache;
84EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030085
Avi Kivity15ad7142007-07-11 18:17:21 +030086static __read_mostly struct preempt_ops kvm_preempt_ops;
87
Hollis Blanchard76f7c872008-04-15 16:05:42 -050088struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080089
Avi Kivitybccf2152007-02-21 18:04:26 +020090static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020092#ifdef CONFIG_COMPAT
93static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 unsigned long arg);
95#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020096static int hardware_enable_all(void);
97static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020098
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020099static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -0800100
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
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +0000110bool kvm_is_reserved_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))
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +0000113 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;
Avi Kivity15ad7142007-07-11 18:17:21 +0300127 cpu = get_cpu();
128 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200129 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300130 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300131 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200132}
133
Carsten Otte313a3dc2007-10-11 19:16:52 +0200134void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135{
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200137 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300138 preempt_notifier_unregister(&vcpu->preempt_notifier);
139 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800140 mutex_unlock(&vcpu->mutex);
141}
142
Avi Kivityd9e368d2007-06-07 19:18:30 +0300143static void ack_flush(void *_completed)
144{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300145}
146
Tang Chen445b8232014-09-24 15:57:55 +0800147bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300148{
Avi Kivity597a5f52008-07-20 14:24:22 +0300149 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030150 cpumask_var_t cpus;
151 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300152 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300153
Li Zefan79f55992009-06-15 14:58:26 +0800154 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030155
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800156 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300157 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800158 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300159 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800160
161 /* Set ->requests bit before we read ->mode */
162 smp_mb();
163
164 if (cpus != NULL && cpu != -1 && cpu != me &&
165 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030166 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300167 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030168 if (unlikely(cpus == NULL))
169 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
170 else if (!cpumask_empty(cpus))
171 smp_call_function_many(cpus, ack_flush, NULL, 1);
172 else
173 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800174 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030175 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030176 return called;
177}
178
179void kvm_flush_remote_tlbs(struct kvm *kvm)
180{
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800181 long dirty_count = kvm->tlbs_dirty;
182
183 smp_mb();
Tang Chen445b8232014-09-24 15:57:55 +0800184 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
Rusty Russell49846892008-12-08 20:26:24 +1030185 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800186 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300187}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530188EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300189
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500190void kvm_reload_remote_mmus(struct kvm *kvm)
191{
Tang Chen445b8232014-09-24 15:57:55 +0800192 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500193}
194
Marcelo Tosattid8281992012-11-27 23:29:01 -0200195void kvm_make_mclock_inprogress_request(struct kvm *kvm)
196{
Tang Chen445b8232014-09-24 15:57:55 +0800197 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
Marcelo Tosattid8281992012-11-27 23:29:01 -0200198}
199
Yang Zhang3d81bc72013-04-11 19:25:13 +0800200void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800201{
Tang Chen445b8232014-09-24 15:57:55 +0800202 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800203}
204
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000205int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
206{
207 struct page *page;
208 int r;
209
210 mutex_init(&vcpu->mutex);
211 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000212 vcpu->kvm = kvm;
213 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500214 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300215 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200216 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000217
218 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
219 if (!page) {
220 r = -ENOMEM;
221 goto fail;
222 }
223 vcpu->run = page_address(page);
224
Raghavendra K T4c088492012-07-18 19:07:46 +0530225 kvm_vcpu_set_in_spin_loop(vcpu, false);
226 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530227 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530228
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800229 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000230 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800231 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232 return 0;
233
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000234fail_free_run:
235 free_page((unsigned long)vcpu->run);
236fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000237 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000238}
239EXPORT_SYMBOL_GPL(kvm_vcpu_init);
240
241void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
242{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500243 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800244 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000245 free_page((unsigned long)vcpu->run);
246}
247EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
248
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200249#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
250static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
251{
252 return container_of(mn, struct kvm, mmu_notifier);
253}
254
255static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
256 struct mm_struct *mm,
257 unsigned long address)
258{
259 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200260 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200261
262 /*
263 * When ->invalidate_page runs, the linux pte has been zapped
264 * already but the page is still allocated until
265 * ->invalidate_page returns. So if we increase the sequence
266 * here the kvm page fault will notice if the spte can't be
267 * established because the page is going to be freed. If
268 * instead the kvm page fault establishes the spte before
269 * ->invalidate_page runs, kvm_unmap_hva will release it
270 * before returning.
271 *
272 * The sequence increase only need to be seen at spin_unlock
273 * time, and not at spin_lock time.
274 *
275 * Increasing the sequence after the spin_unlock would be
276 * unsafe because the kvm page fault could then establish the
277 * pte after kvm_unmap_hva returned, without noticing the page
278 * is going to be freed.
279 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200280 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200281 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900282
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200283 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800284 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200285 /* we've to flush the tlb before the pages can be freed */
286 if (need_tlb_flush)
287 kvm_flush_remote_tlbs(kvm);
288
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900289 spin_unlock(&kvm->mmu_lock);
Tang Chenfe715572014-09-24 15:57:57 +0800290
291 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
292
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900293 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200294}
295
Izik Eidus3da0dd42009-09-23 21:47:18 +0300296static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
297 struct mm_struct *mm,
298 unsigned long address,
299 pte_t pte)
300{
301 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200302 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300303
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200304 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300305 spin_lock(&kvm->mmu_lock);
306 kvm->mmu_notifier_seq++;
307 kvm_set_spte_hva(kvm, address, pte);
308 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200309 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300310}
311
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200312static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
313 struct mm_struct *mm,
314 unsigned long start,
315 unsigned long end)
316{
317 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200318 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200319
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200320 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200321 spin_lock(&kvm->mmu_lock);
322 /*
323 * The count increase must become visible at unlock time as no
324 * spte can be established without taking the mmu_lock and
325 * count is also read inside the mmu_lock critical section.
326 */
327 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900328 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800329 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200330 /* we've to flush the tlb before the pages can be freed */
331 if (need_tlb_flush)
332 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900333
334 spin_unlock(&kvm->mmu_lock);
335 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200336}
337
338static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
339 struct mm_struct *mm,
340 unsigned long start,
341 unsigned long end)
342{
343 struct kvm *kvm = mmu_notifier_to_kvm(mn);
344
345 spin_lock(&kvm->mmu_lock);
346 /*
347 * This sequence increase will notify the kvm page fault that
348 * the page that is going to be mapped in the spte could have
349 * been freed.
350 */
351 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000352 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200353 /*
354 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000355 * below count decrease, which is ensured by the smp_wmb above
356 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200357 */
358 kvm->mmu_notifier_count--;
359 spin_unlock(&kvm->mmu_lock);
360
361 BUG_ON(kvm->mmu_notifier_count < 0);
362}
363
364static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
365 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700366 unsigned long start,
367 unsigned long end)
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200368{
369 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200370 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200371
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200372 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200373 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200374
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700375 young = kvm_age_hva(kvm, start, end);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200376 if (young)
377 kvm_flush_remote_tlbs(kvm);
378
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900379 spin_unlock(&kvm->mmu_lock);
380 srcu_read_unlock(&kvm->srcu, idx);
381
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200382 return young;
383}
384
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800385static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
386 struct mm_struct *mm,
387 unsigned long address)
388{
389 struct kvm *kvm = mmu_notifier_to_kvm(mn);
390 int young, idx;
391
392 idx = srcu_read_lock(&kvm->srcu);
393 spin_lock(&kvm->mmu_lock);
394 young = kvm_test_age_hva(kvm, address);
395 spin_unlock(&kvm->mmu_lock);
396 srcu_read_unlock(&kvm->srcu, idx);
397
398 return young;
399}
400
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100401static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
402 struct mm_struct *mm)
403{
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800405 int idx;
406
407 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300408 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800409 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100410}
411
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200412static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
413 .invalidate_page = kvm_mmu_notifier_invalidate_page,
414 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
415 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
416 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800417 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300418 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100419 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200420};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200421
422static int kvm_init_mmu_notifier(struct kvm *kvm)
423{
424 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
425 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
426}
427
428#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
429
430static int kvm_init_mmu_notifier(struct kvm *kvm)
431{
432 return 0;
433}
434
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200435#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
436
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800437static void kvm_init_memslots_id(struct kvm *kvm)
438{
439 int i;
440 struct kvm_memslots *slots = kvm->memslots;
441
442 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800443 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800444}
445
Carsten Ottee08b9632012-01-04 10:25:20 +0100446static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800447{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100448 int r, i;
449 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800450
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100451 if (!kvm)
452 return ERR_PTR(-ENOMEM);
453
Carsten Ottee08b9632012-01-04 10:25:20 +0100454 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100455 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100456 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200457
458 r = hardware_enable_all();
459 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100460 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200461
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200462#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300463 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200464#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800465
Alex Williamson1e702d92012-12-10 10:33:32 -0700466 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
467
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200468 r = -ENOMEM;
469 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
470 if (!kvm->memslots)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100471 goto out_err_no_srcu;
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200472
473 /*
474 * Init kvm generation close to the maximum to easily test the
475 * code of handling generation number wrap-around.
476 */
477 kvm->memslots->generation = -150;
478
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800479 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200480 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100481 goto out_err_no_srcu;
482 if (init_srcu_struct(&kvm->irq_srcu))
483 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
486 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100487 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200488 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200489 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700490
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900491 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400494 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800495 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300496 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200497 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300498 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000499 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700500
501 r = kvm_init_mmu_notifier(kvm);
502 if (r)
503 goto out_err;
504
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200505 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000506 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200507 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100508
Avi Kivityf17abe92007-02-21 19:28:04 +0200509 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200510
511out_err:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100512 cleanup_srcu_struct(&kvm->irq_srcu);
513out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100514 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100515out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200516 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100517out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200518 for (i = 0; i < KVM_NR_BUSES; i++)
519 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200520 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100521 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200522 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200523}
524
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900525/*
526 * Avoid using vmalloc for a small buffer.
527 * Should not be used when the size is statically known.
528 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900529void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900530{
531 if (size > PAGE_SIZE)
532 return vzalloc(size);
533 else
534 return kzalloc(size, GFP_KERNEL);
535}
536
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900537void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900538{
539 if (is_vmalloc_addr(addr))
540 vfree(addr);
541 else
542 kfree(addr);
543}
544
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900545static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
546{
547 if (!memslot->dirty_bitmap)
548 return;
549
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900550 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900551 memslot->dirty_bitmap = NULL;
552}
553
Avi Kivity6aa8b732006-12-10 02:21:36 -0800554/*
555 * Free any memory in @free but not in @dont.
556 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530557static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800558 struct kvm_memory_slot *dont)
559{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900561 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800562
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530563 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300564
Avi Kivity6aa8b732006-12-10 02:21:36 -0800565 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566}
567
Stephen Hemminger79408762013-12-29 12:12:29 -0800568static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200570 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800571 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800572
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800573 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530574 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200575
576 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800577}
578
Scott Wood07f0a7b2013-04-25 14:11:23 +0000579static void kvm_destroy_devices(struct kvm *kvm)
580{
581 struct list_head *node, *tmp;
582
583 list_for_each_safe(node, tmp, &kvm->devices) {
584 struct kvm_device *dev =
585 list_entry(node, struct kvm_device, vm_node);
586
587 list_del(node);
588 dev->ops->destroy(dev);
589 }
590}
591
Avi Kivityf17abe92007-02-21 19:28:04 +0200592static void kvm_destroy_vm(struct kvm *kvm)
593{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200594 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200595 struct mm_struct *mm = kvm->mm;
596
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800597 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200598 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800599 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200600 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200601 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200602 for (i = 0; i < KVM_NR_BUSES; i++)
603 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200604 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200605#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
606 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200607#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300608 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200609#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800610 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000611 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100612 kvm_free_physmem(kvm);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200613 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100614 cleanup_srcu_struct(&kvm->srcu);
615 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200616 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200617 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200618}
619
Izik Eidusd39f13b2008-03-30 16:01:25 +0300620void kvm_get_kvm(struct kvm *kvm)
621{
622 atomic_inc(&kvm->users_count);
623}
624EXPORT_SYMBOL_GPL(kvm_get_kvm);
625
626void kvm_put_kvm(struct kvm *kvm)
627{
628 if (atomic_dec_and_test(&kvm->users_count))
629 kvm_destroy_vm(kvm);
630}
631EXPORT_SYMBOL_GPL(kvm_put_kvm);
632
633
Avi Kivityf17abe92007-02-21 19:28:04 +0200634static int kvm_vm_release(struct inode *inode, struct file *filp)
635{
636 struct kvm *kvm = filp->private_data;
637
Gregory Haskins721eecb2009-05-20 10:30:49 -0400638 kvm_irqfd_release(kvm);
639
Izik Eidusd39f13b2008-03-30 16:01:25 +0300640 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800641 return 0;
642}
643
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900644/*
645 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900646 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900647 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900648static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
649{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900650 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900651
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900652 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900653 if (!memslot->dirty_bitmap)
654 return -ENOMEM;
655
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900656 return 0;
657}
658
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800659/*
Igor Mammedov0e60b072014-12-01 17:29:26 +0000660 * Insert memslot and re-sort memslots based on their GFN,
661 * so binary search could be used to lookup GFN.
662 * Sorting algorithm takes advantage of having initially
663 * sorted array and known changed memslot position.
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800664 */
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100665static void update_memslots(struct kvm_memslots *slots,
666 struct kvm_memory_slot *new)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800667{
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100668 int id = new->id;
669 int i = slots->id_to_index[id];
Igor Mammedov063584d2014-11-13 23:00:13 +0000670 struct kvm_memory_slot *mslots = slots->memslots;
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800671
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100672 WARN_ON(mslots[i].id != id);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000673 if (!new->npages) {
Igor Mammedov0e60b072014-12-01 17:29:26 +0000674 new->base_gfn = 0;
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000675 if (mslots[i].npages)
676 slots->used_slots--;
677 } else {
678 if (!mslots[i].npages)
679 slots->used_slots++;
680 }
Igor Mammedov0e60b072014-12-01 17:29:26 +0000681
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000682 while (i < KVM_MEM_SLOTS_NUM - 1 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000683 new->base_gfn <= mslots[i + 1].base_gfn) {
684 if (!mslots[i + 1].npages)
685 break;
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000686 mslots[i] = mslots[i + 1];
687 slots->id_to_index[mslots[i].id] = i;
688 i++;
689 }
690 while (i > 0 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000691 new->base_gfn > mslots[i - 1].base_gfn) {
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000692 mslots[i] = mslots[i - 1];
693 slots->id_to_index[mslots[i].id] = i;
694 i--;
Igor Mammedov063584d2014-11-13 23:00:13 +0000695 }
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800696
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100697 mslots[i] = *new;
698 slots->id_to_index[mslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800699}
700
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800701static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
702{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800703 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
704
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200705#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800706 valid_flags |= KVM_MEM_READONLY;
707#endif
708
709 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800710 return -EINVAL;
711
712 return 0;
713}
714
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200715static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100716 struct kvm_memslots *slots)
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200717{
718 struct kvm_memslots *old_memslots = kvm->memslots;
719
David Matlackee3d1572014-08-18 15:46:06 -0700720 /*
721 * Set the low bit in the generation, which disables SPTE caching
722 * until the end of synchronize_srcu_expedited.
723 */
724 WARN_ON(old_memslots->generation & 1);
725 slots->generation = old_memslots->generation + 1;
726
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200727 rcu_assign_pointer(kvm->memslots, slots);
728 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900729
David Matlackee3d1572014-08-18 15:46:06 -0700730 /*
731 * Increment the new memslot generation a second time. This prevents
732 * vm exits that race with memslot updates from caching a memslot
733 * generation that will (potentially) be valid forever.
734 */
735 slots->generation++;
736
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900737 kvm_arch_memslots_updated(kvm);
738
739 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200740}
741
Avi Kivity6aa8b732006-12-10 02:21:36 -0800742/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800743 * Allocate some memory and give it an address in the guest physical address
744 * space.
745 *
746 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800747 *
Dominik Dingel02d5d552014-10-27 16:22:56 +0100748 * Must be called holding kvm->slots_lock for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800749 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800750int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900751 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800752{
Avi Kivity8234b222010-12-27 12:08:45 +0200753 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800754 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200755 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900756 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800757 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700758 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900759 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800760
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800761 r = check_memory_region_flags(mem);
762 if (r)
763 goto out;
764
Avi Kivity6aa8b732006-12-10 02:21:36 -0800765 r = -EINVAL;
766 /* General sanity checks */
767 if (mem->memory_size & (PAGE_SIZE - 1))
768 goto out;
769 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
770 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900771 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900772 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900773 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200774 !access_ok(VERIFY_WRITE,
775 (void __user *)(unsigned long)mem->userspace_addr,
776 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600777 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800778 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779 goto out;
780 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
781 goto out;
782
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900783 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800784 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
785 npages = mem->memory_size >> PAGE_SHIFT;
786
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900787 if (npages > KVM_MEM_MAX_NR_PAGES)
788 goto out;
789
Avi Kivity6aa8b732006-12-10 02:21:36 -0800790 if (!npages)
791 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
792
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900793 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800794
Avi Kivitye36d96f2010-06-21 10:56:36 +0300795 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800796 new.base_gfn = base_gfn;
797 new.npages = npages;
798 new.flags = mem->flags;
799
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900800 if (npages) {
801 if (!old.npages)
802 change = KVM_MR_CREATE;
803 else { /* Modify an existing slot. */
804 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900805 (npages != old.npages) ||
806 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900807 goto out;
808
809 if (base_gfn != old.base_gfn)
810 change = KVM_MR_MOVE;
811 else if (new.flags != old.flags)
812 change = KVM_MR_FLAGS_ONLY;
813 else { /* Nothing to change. */
814 r = 0;
815 goto out;
816 }
817 }
818 } else if (old.npages) {
819 change = KVM_MR_DELETE;
820 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900821 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800822
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900823 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900824 /* Check for overlaps */
825 r = -EEXIST;
826 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900827 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
828 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900829 continue;
830 if (!((base_gfn + npages <= slot->base_gfn) ||
831 (base_gfn >= slot->base_gfn + slot->npages)))
832 goto out;
833 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800835
Avi Kivity6aa8b732006-12-10 02:21:36 -0800836 /* Free page dirty bitmap if unneeded */
837 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000838 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800839
840 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900841 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900842 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900843
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530844 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900845 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800846 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200847
Avi Kivity6aa8b732006-12-10 02:21:36 -0800848 /* Allocate page dirty bitmap if needed */
849 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900850 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800851 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800852 }
853
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100854 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
855 GFP_KERNEL);
856 if (!slots)
857 goto out_free;
858
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900859 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Xiao Guangrong28a37542011-11-24 19:04:35 +0800860 slot = id_to_memslot(slots, mem->slot);
861 slot->flags |= KVM_MEMSLOT_INVALID;
862
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100863 old_memslots = install_new_memslots(kvm, slots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200864
Alex Williamsone40f1932012-12-10 10:32:57 -0700865 /* slot was deleted or moved, clear iommu mapping */
866 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300867 /* From this point no new shadow pages pointing to a deleted,
868 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200869 *
870 * validation of sp->gfn happens in:
871 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
872 * - kvm_is_visible_gfn (mmu_check_roots)
873 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300874 kvm_arch_flush_shadow_memslot(kvm, slot);
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100875
876 /*
877 * We can re-use the old_memslots from above, the only difference
878 * from the currently installed memslots is the invalid flag. This
879 * will get overwritten by update_memslots anyway.
880 */
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700881 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200882 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300883
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900884 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200885 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700886 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200887
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200888 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900889 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200890 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900891 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200892 }
893
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100894 update_memslots(slots, &new);
895 old_memslots = install_new_memslots(kvm, slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800896
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900897 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800898
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530899 kvm_free_physmem_slot(kvm, &old, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200900 kfree(old_memslots);
901
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200902 /*
903 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800904 * un-mapped and re-mapped if their base changes. Since base change
905 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800906 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800907 * slots (size changes, userspace addr changes and read-only flag
908 * changes) is disallowed above, so any other attribute changes getting
909 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200910 */
911 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800912 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800913 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800914 }
915
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916 return 0;
917
918out_slots:
919 kfree(slots);
920out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530921 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800922out:
923 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200924}
925EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800926
927int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900928 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800929{
930 int r;
931
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200932 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900933 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200934 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800935 return r;
936}
Izik Eidus210c7c42007-10-24 23:52:57 +0200937EXPORT_SYMBOL_GPL(kvm_set_memory_region);
938
Stephen Hemminger79408762013-12-29 12:12:29 -0800939static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
940 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200941{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700942 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200943 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900944 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945}
946
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800947int kvm_get_dirty_log(struct kvm *kvm,
948 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800949{
950 struct kvm_memory_slot *memslot;
951 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900952 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800953 unsigned long any = 0;
954
Avi Kivity6aa8b732006-12-10 02:21:36 -0800955 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700956 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957 goto out;
958
Xiao Guangrong28a37542011-11-24 19:04:35 +0800959 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800960 r = -ENOENT;
961 if (!memslot->dirty_bitmap)
962 goto out;
963
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900964 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965
Uri Lublincd1a4a92007-02-22 16:43:09 +0200966 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967 any = memslot->dirty_bitmap[i];
968
969 r = -EFAULT;
970 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
971 goto out;
972
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800973 if (any)
974 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800975
976 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800978 return r;
979}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530980EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800981
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900982bool kvm_largepages_enabled(void)
983{
984 return largepages_enabled;
985}
986
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300987void kvm_disable_largepages(void)
988{
989 largepages_enabled = false;
990}
991EXPORT_SYMBOL_GPL(kvm_disable_largepages);
992
Gleb Natapov49c77542010-10-18 15:22:23 +0200993struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
994{
995 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
996}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300997EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800998
Izik Eiduse0d62c72007-10-24 23:57:46 +0200999int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1000{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001001 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001002
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001003 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001004 memslot->flags & KVM_MEMSLOT_INVALID)
1005 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001006
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001007 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001008}
1009EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1010
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001011unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1012{
1013 struct vm_area_struct *vma;
1014 unsigned long addr, size;
1015
1016 size = PAGE_SIZE;
1017
1018 addr = gfn_to_hva(kvm, gfn);
1019 if (kvm_is_error_hva(addr))
1020 return PAGE_SIZE;
1021
1022 down_read(&current->mm->mmap_sem);
1023 vma = find_vma(current->mm, addr);
1024 if (!vma)
1025 goto out;
1026
1027 size = vma_kernel_pagesize(vma);
1028
1029out:
1030 up_read(&current->mm->mmap_sem);
1031
1032 return size;
1033}
1034
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001035static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1036{
1037 return slot->flags & KVM_MEM_READONLY;
1038}
1039
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001040static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1041 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001042{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001043 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001044 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001045
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001046 if (memslot_is_readonly(slot) && write)
1047 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001048
1049 if (nr_pages)
1050 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1051
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001052 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001053}
Xiao Guangrong48987782010-08-22 19:11:43 +08001054
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056 gfn_t *nr_pages)
1057{
1058 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1059}
1060
1061unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001062 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001063{
1064 return gfn_to_hva_many(slot, gfn, NULL);
1065}
1066EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1067
Xiao Guangrong48987782010-08-22 19:11:43 +08001068unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1069{
Gleb Natapov49c77542010-10-18 15:22:23 +02001070 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001071}
Sheng Yang0d150292008-04-25 21:44:50 +08001072EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001073
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001074/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001075 * If writable is set to false, the hva returned by this function is only
1076 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001077 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001078unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1079 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001080{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001081 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1082
1083 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001084 *writable = !memslot_is_readonly(slot);
1085
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001086 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001087}
1088
Christoffer Dall64d83122014-08-19 12:15:00 +02001089unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1090{
1091 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1092
1093 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1094}
1095
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001096static int kvm_read_hva(void *data, void __user *hva, int len)
1097{
1098 return __copy_from_user(data, hva, len);
1099}
1100
1101static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1102{
1103 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001104}
1105
Geoff Levand39369f72013-04-05 19:20:30 +00001106static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001107 unsigned long start, int write, struct page **page)
1108{
1109 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1110
1111 if (write)
1112 flags |= FOLL_WRITE;
1113
1114 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1115}
1116
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001117int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1118 unsigned long addr, bool write_fault,
1119 struct page **pagep)
1120{
1121 int npages;
1122 int locked = 1;
1123 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1124 (pagep ? FOLL_GET : 0) |
1125 (write_fault ? FOLL_WRITE : 0);
1126
1127 /*
1128 * If retrying the fault, we get here *not* having allowed the filemap
1129 * to wait on the page lock. We should now allow waiting on the IO with
1130 * the mmap semaphore released.
1131 */
1132 down_read(&mm->mmap_sem);
1133 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1134 &locked);
1135 if (!locked) {
Andres Lagar-Cavillabb0ca6a2014-09-25 15:26:50 -07001136 VM_BUG_ON(npages);
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001137
1138 if (!pagep)
1139 return 0;
1140
1141 /*
1142 * The previous call has now waited on the IO. Now we can
1143 * retry and complete. Pass TRIED to ensure we do not re
1144 * schedule async IO (see e.g. filemap_fault).
1145 */
1146 down_read(&mm->mmap_sem);
1147 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1148 pagep, NULL, NULL);
1149 }
1150 up_read(&mm->mmap_sem);
1151 return npages;
1152}
1153
Huang Yingfafc3db2011-01-30 11:15:49 +08001154static inline int check_user_page_hwpoison(unsigned long addr)
1155{
1156 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1157
1158 rc = __get_user_pages(current, current->mm, addr, 1,
1159 flags, NULL, NULL, NULL);
1160 return rc == -EHWPOISON;
1161}
1162
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001163/*
1164 * The atomic path to get the writable pfn which will be stored in @pfn,
1165 * true indicates success, otherwise false is returned.
1166 */
1167static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1168 bool write_fault, bool *writable, pfn_t *pfn)
1169{
1170 struct page *page[1];
1171 int npages;
1172
1173 if (!(async || atomic))
1174 return false;
1175
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001176 /*
1177 * Fast pin a writable pfn only if it is a write fault request
1178 * or the caller allows to map a writable pfn for a read fault
1179 * request.
1180 */
1181 if (!(write_fault || writable))
1182 return false;
1183
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001184 npages = __get_user_pages_fast(addr, 1, 1, page);
1185 if (npages == 1) {
1186 *pfn = page_to_pfn(page[0]);
1187
1188 if (writable)
1189 *writable = true;
1190 return true;
1191 }
1192
1193 return false;
1194}
1195
1196/*
1197 * The slow path to get the pfn of the specified host virtual address,
1198 * 1 indicates success, -errno is returned if error is detected.
1199 */
1200static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1201 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001202{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001203 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001204 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001205
1206 might_sleep();
1207
1208 if (writable)
1209 *writable = write_fault;
1210
1211 if (async) {
1212 down_read(&current->mm->mmap_sem);
1213 npages = get_user_page_nowait(current, current->mm,
1214 addr, write_fault, page);
1215 up_read(&current->mm->mmap_sem);
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001216 } else {
1217 /*
1218 * By now we have tried gup_fast, and possibly async_pf, and we
1219 * are certainly not atomic. Time to retry the gup, allowing
1220 * mmap semaphore to be relinquished in the case of IO.
1221 */
1222 npages = kvm_get_user_page_io(current, current->mm, addr,
1223 write_fault, page);
1224 }
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001225 if (npages != 1)
1226 return npages;
1227
1228 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001229 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001230 struct page *wpage[1];
1231
1232 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1233 if (npages == 1) {
1234 *writable = true;
1235 put_page(page[0]);
1236 page[0] = wpage[0];
1237 }
1238
1239 npages = 1;
1240 }
1241 *pfn = page_to_pfn(page[0]);
1242 return npages;
1243}
1244
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001245static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1246{
1247 if (unlikely(!(vma->vm_flags & VM_READ)))
1248 return false;
1249
1250 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1251 return false;
1252
1253 return true;
1254}
1255
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001256/*
1257 * Pin guest page in memory and return its pfn.
1258 * @addr: host virtual address which maps memory to the guest
1259 * @atomic: whether this function can sleep
1260 * @async: whether this function need to wait IO complete if the
1261 * host page is not in the memory
1262 * @write_fault: whether we should get a writable host page
1263 * @writable: whether it allows to map a writable host page for !@write_fault
1264 *
1265 * The function will map a writable host page for these two cases:
1266 * 1): @write_fault = true
1267 * 2): @write_fault = false && @writable, @writable will tell the caller
1268 * whether the mapping is writable.
1269 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001270static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1271 bool write_fault, bool *writable)
1272{
1273 struct vm_area_struct *vma;
1274 pfn_t pfn = 0;
1275 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001276
Gleb Natapovaf585b92010-10-14 11:22:46 +02001277 /* we can do it either atomically or asynchronously, not both */
1278 BUG_ON(atomic && async);
1279
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001280 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1281 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001282
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001283 if (atomic)
1284 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001285
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001286 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1287 if (npages == 1)
1288 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001289
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001290 down_read(&current->mm->mmap_sem);
1291 if (npages == -EHWPOISON ||
1292 (!async && check_user_page_hwpoison(addr))) {
1293 pfn = KVM_PFN_ERR_HWPOISON;
1294 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001295 }
Izik Eidus539cb662007-11-11 22:05:04 +02001296
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001297 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001298
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001299 if (vma == NULL)
1300 pfn = KVM_PFN_ERR_FAULT;
1301 else if ((vma->vm_flags & VM_PFNMAP)) {
1302 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1303 vma->vm_pgoff;
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001304 BUG_ON(!kvm_is_reserved_pfn(pfn));
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001305 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001306 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001307 *async = true;
1308 pfn = KVM_PFN_ERR_FAULT;
1309 }
1310exit:
1311 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001312 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001313}
1314
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001315static pfn_t
1316__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1317 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001318{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001319 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1320
1321 if (addr == KVM_HVA_ERR_RO_BAD)
1322 return KVM_PFN_ERR_RO_FAULT;
1323
1324 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001325 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001326
1327 /* Do not map writable pfn in the readonly memslot. */
1328 if (writable && memslot_is_readonly(slot)) {
1329 *writable = false;
1330 writable = NULL;
1331 }
1332
1333 return hva_to_pfn(addr, atomic, async, write_fault,
1334 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001335}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001336
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001337static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1338 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001339{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001340 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001341
Gleb Natapovaf585b92010-10-14 11:22:46 +02001342 if (async)
1343 *async = false;
1344
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001345 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001346
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001347 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1348 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001349}
1350
1351pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1352{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001353 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001354}
1355EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1356
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001357pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1358 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001359{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001360 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001361}
1362EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1363
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001364pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1365{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001366 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001367}
Anthony Liguori35149e22008-04-02 14:46:56 -05001368EXPORT_SYMBOL_GPL(gfn_to_pfn);
1369
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001370pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1371 bool *writable)
1372{
1373 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1374}
1375EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1376
Xiao Guangrongd5661042012-07-17 21:56:16 +08001377pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001378{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001379 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001380}
1381
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001382pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1383{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001384 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001385}
1386EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1387
Xiao Guangrong48987782010-08-22 19:11:43 +08001388int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1389 int nr_pages)
1390{
1391 unsigned long addr;
1392 gfn_t entry;
1393
Gleb Natapov49c77542010-10-18 15:22:23 +02001394 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001395 if (kvm_is_error_hva(addr))
1396 return -1;
1397
1398 if (entry < nr_pages)
1399 return 0;
1400
1401 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1402}
1403EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1404
Xiao Guangronga2766322012-07-26 11:58:59 +08001405static struct page *kvm_pfn_to_page(pfn_t pfn)
1406{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001407 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001408 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001409
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001410 if (kvm_is_reserved_pfn(pfn)) {
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001411 WARN_ON(1);
1412 return KVM_ERR_PTR_BAD_PAGE;
1413 }
1414
Xiao Guangronga2766322012-07-26 11:58:59 +08001415 return pfn_to_page(pfn);
1416}
1417
Anthony Liguori35149e22008-04-02 14:46:56 -05001418struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1419{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001420 pfn_t pfn;
1421
1422 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001423
Xiao Guangronga2766322012-07-26 11:58:59 +08001424 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001425}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001426
Avi Kivity954bbbc2007-03-30 14:02:32 +03001427EXPORT_SYMBOL_GPL(gfn_to_page);
1428
Izik Eidusb4231d62007-11-20 11:49:33 +02001429void kvm_release_page_clean(struct page *page)
1430{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001431 WARN_ON(is_error_page(page));
1432
Anthony Liguori35149e22008-04-02 14:46:56 -05001433 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001434}
1435EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1436
Anthony Liguori35149e22008-04-02 14:46:56 -05001437void kvm_release_pfn_clean(pfn_t pfn)
1438{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001439 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001440 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001441}
1442EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1443
Izik Eidusb4231d62007-11-20 11:49:33 +02001444void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001445{
Xiao Guangronga2766322012-07-26 11:58:59 +08001446 WARN_ON(is_error_page(page));
1447
Anthony Liguori35149e22008-04-02 14:46:56 -05001448 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001449}
Izik Eidusb4231d62007-11-20 11:49:33 +02001450EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001451
Stephen Hemminger79408762013-12-29 12:12:29 -08001452static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001453{
1454 kvm_set_pfn_dirty(pfn);
1455 kvm_release_pfn_clean(pfn);
1456}
Anthony Liguori35149e22008-04-02 14:46:56 -05001457
1458void kvm_set_pfn_dirty(pfn_t pfn)
1459{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001460 if (!kvm_is_reserved_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001461 struct page *page = pfn_to_page(pfn);
1462 if (!PageReserved(page))
1463 SetPageDirty(page);
1464 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001465}
1466EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1467
1468void kvm_set_pfn_accessed(pfn_t pfn)
1469{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001470 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001471 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001472}
1473EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1474
1475void kvm_get_pfn(pfn_t pfn)
1476{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001477 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001478 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001479}
1480EXPORT_SYMBOL_GPL(kvm_get_pfn);
1481
Izik Eidus195aefd2007-10-01 22:14:18 +02001482static int next_segment(unsigned long len, int offset)
1483{
1484 if (len > PAGE_SIZE - offset)
1485 return PAGE_SIZE - offset;
1486 else
1487 return len;
1488}
1489
1490int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1491 int len)
1492{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001493 int r;
1494 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001495
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001496 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001497 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001498 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001499 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001500 if (r)
1501 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001502 return 0;
1503}
1504EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1505
1506int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, 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_read_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}
1524EXPORT_SYMBOL_GPL(kvm_read_guest);
1525
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001526int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1527 unsigned long len)
1528{
1529 int r;
1530 unsigned long addr;
1531 gfn_t gfn = gpa >> PAGE_SHIFT;
1532 int offset = offset_in_page(gpa);
1533
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001534 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001535 if (kvm_is_error_hva(addr))
1536 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001537 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001538 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001539 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001540 if (r)
1541 return -EFAULT;
1542 return 0;
1543}
1544EXPORT_SYMBOL(kvm_read_guest_atomic);
1545
Izik Eidus195aefd2007-10-01 22:14:18 +02001546int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1547 int offset, int len)
1548{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001549 int r;
1550 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001551
Izik Eiduse0506bc2007-11-11 22:10:22 +02001552 addr = gfn_to_hva(kvm, gfn);
1553 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001554 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001555 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001556 if (r)
1557 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001558 mark_page_dirty(kvm, gfn);
1559 return 0;
1560}
1561EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1562
1563int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1564 unsigned long len)
1565{
1566 gfn_t gfn = gpa >> PAGE_SHIFT;
1567 int seg;
1568 int offset = offset_in_page(gpa);
1569 int ret;
1570
1571 while ((seg = next_segment(len, offset)) != 0) {
1572 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1573 if (ret < 0)
1574 return ret;
1575 offset = 0;
1576 len -= seg;
1577 data += seg;
1578 ++gfn;
1579 }
1580 return 0;
1581}
1582
Gleb Natapov49c77542010-10-18 15:22:23 +02001583int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001584 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001585{
1586 struct kvm_memslots *slots = kvm_memslots(kvm);
1587 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001588 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1589 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1590 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1591 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001592
1593 ghc->gpa = gpa;
1594 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001595 ghc->len = len;
1596 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1597 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1598 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001599 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001600 } else {
1601 /*
1602 * If the requested region crosses two memslots, we still
1603 * verify that the entire region is valid here.
1604 */
1605 while (start_gfn <= end_gfn) {
1606 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1607 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1608 &nr_pages_avail);
1609 if (kvm_is_error_hva(ghc->hva))
1610 return -EFAULT;
1611 start_gfn += nr_pages_avail;
1612 }
1613 /* Use the slow path for cross page reads and writes. */
1614 ghc->memslot = NULL;
1615 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001616 return 0;
1617}
1618EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1619
1620int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1621 void *data, unsigned long len)
1622{
1623 struct kvm_memslots *slots = kvm_memslots(kvm);
1624 int r;
1625
Andrew Honig8f964522013-03-29 09:35:21 -07001626 BUG_ON(len > ghc->len);
1627
Gleb Natapov49c77542010-10-18 15:22:23 +02001628 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001629 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1630
1631 if (unlikely(!ghc->memslot))
1632 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001633
1634 if (kvm_is_error_hva(ghc->hva))
1635 return -EFAULT;
1636
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001637 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001638 if (r)
1639 return -EFAULT;
1640 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1641
1642 return 0;
1643}
1644EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1645
Gleb Natapove03b6442011-07-11 15:28:11 -04001646int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1647 void *data, unsigned long len)
1648{
1649 struct kvm_memslots *slots = kvm_memslots(kvm);
1650 int r;
1651
Andrew Honig8f964522013-03-29 09:35:21 -07001652 BUG_ON(len > ghc->len);
1653
Gleb Natapove03b6442011-07-11 15:28:11 -04001654 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001655 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1656
1657 if (unlikely(!ghc->memslot))
1658 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001659
1660 if (kvm_is_error_hva(ghc->hva))
1661 return -EFAULT;
1662
1663 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1664 if (r)
1665 return -EFAULT;
1666
1667 return 0;
1668}
1669EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1670
Izik Eidus195aefd2007-10-01 22:14:18 +02001671int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1672{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001673 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1674
1675 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001676}
1677EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1678
1679int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1680{
1681 gfn_t gfn = gpa >> PAGE_SHIFT;
1682 int seg;
1683 int offset = offset_in_page(gpa);
1684 int ret;
1685
1686 while ((seg = next_segment(len, offset)) != 0) {
1687 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1688 if (ret < 0)
1689 return ret;
1690 offset = 0;
1691 len -= seg;
1692 ++gfn;
1693 }
1694 return 0;
1695}
1696EXPORT_SYMBOL_GPL(kvm_clear_guest);
1697
Stephen Hemminger79408762013-12-29 12:12:29 -08001698static void mark_page_dirty_in_slot(struct kvm *kvm,
1699 struct kvm_memory_slot *memslot,
1700 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001701{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001702 if (memslot && memslot->dirty_bitmap) {
1703 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001704
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001705 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001706 }
1707}
1708
Gleb Natapov49c77542010-10-18 15:22:23 +02001709void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1710{
1711 struct kvm_memory_slot *memslot;
1712
1713 memslot = gfn_to_memslot(kvm, gfn);
1714 mark_page_dirty_in_slot(kvm, memslot, gfn);
1715}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301716EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001717
Eddie Dongb6958ce2007-07-18 12:15:21 +03001718/*
1719 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1720 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001721void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001722{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001723 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001724
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001725 for (;;) {
1726 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001727
Gleb Natapova1b37102009-07-09 15:33:52 +03001728 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001729 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001730 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001731 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001732 if (kvm_cpu_has_pending_timer(vcpu))
1733 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001734 if (signal_pending(current))
1735 break;
1736
Eddie Dongb6958ce2007-07-18 12:15:21 +03001737 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001738 }
1739
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001740 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001741}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301742EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001743
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001744#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001745/*
1746 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1747 */
1748void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1749{
1750 int me;
1751 int cpu = vcpu->cpu;
1752 wait_queue_head_t *wqp;
1753
1754 wqp = kvm_arch_vcpu_wq(vcpu);
1755 if (waitqueue_active(wqp)) {
1756 wake_up_interruptible(wqp);
1757 ++vcpu->stat.halt_wakeup;
1758 }
1759
1760 me = get_cpu();
1761 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1762 if (kvm_arch_vcpu_should_kick(vcpu))
1763 smp_send_reschedule(cpu);
1764 put_cpu();
1765}
Yang Zhanga20ed542013-04-11 19:25:15 +08001766EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001767#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001768
Dan Carpenterfa933842014-05-23 13:20:42 +03001769int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02001770{
1771 struct pid *pid;
1772 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03001773 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001774
1775 rcu_read_lock();
1776 pid = rcu_dereference(target->pid);
1777 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10001778 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001779 rcu_read_unlock();
1780 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301781 return ret;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301782 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001783 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301784
1785 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001786}
1787EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1788
Raghavendra K T06e48c52012-07-19 15:17:52 +05301789/*
1790 * Helper that checks whether a VCPU is eligible for directed yield.
1791 * Most eligible candidate to yield is decided by following heuristics:
1792 *
1793 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1794 * (preempted lock holder), indicated by @in_spin_loop.
1795 * Set at the beiginning and cleared at the end of interception/PLE handler.
1796 *
1797 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1798 * chance last time (mostly it has become eligible now since we have probably
1799 * yielded to lockholder in last iteration. This is done by toggling
1800 * @dy_eligible each time a VCPU checked for eligibility.)
1801 *
1802 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1803 * to preempted lock-holder could result in wrong VCPU selection and CPU
1804 * burning. Giving priority for a potential lock-holder increases lock
1805 * progress.
1806 *
1807 * Since algorithm is based on heuristics, accessing another VCPU data without
1808 * locking does not harm. It may result in trying to yield to same VCPU, fail
1809 * and continue with next VCPU and so on.
1810 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001811static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301812{
Scott Wood4a55dd72014-01-09 18:43:16 -06001813#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301814 bool eligible;
1815
1816 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02001817 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301818
1819 if (vcpu->spin_loop.in_spin_loop)
1820 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1821
1822 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001823#else
1824 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301825#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001826}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301827
Rik van Riel217ece62011-02-01 09:53:28 -05001828void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001829{
Rik van Riel217ece62011-02-01 09:53:28 -05001830 struct kvm *kvm = me->kvm;
1831 struct kvm_vcpu *vcpu;
1832 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1833 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301834 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001835 int pass;
1836 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001837
Raghavendra K T4c088492012-07-18 19:07:46 +05301838 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001839 /*
1840 * We boost the priority of a VCPU that is runnable but not
1841 * currently running, because it got preempted by something
1842 * else and called schedule in __vcpu_run. Hopefully that
1843 * VCPU is holding the lock that we need and will release it.
1844 * We approximate round-robin by starting at the last boosted VCPU.
1845 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301846 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001847 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001848 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001849 i = last_boosted_vcpu;
1850 continue;
1851 } else if (pass && i > last_boosted_vcpu)
1852 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301853 if (!ACCESS_ONCE(vcpu->preempted))
1854 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001855 if (vcpu == me)
1856 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001857 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001858 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301859 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1860 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301861
1862 yielded = kvm_vcpu_yield_to(vcpu);
1863 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001864 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001865 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301866 } else if (yielded < 0) {
1867 try--;
1868 if (!try)
1869 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001870 }
Rik van Riel217ece62011-02-01 09:53:28 -05001871 }
1872 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301873 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301874
1875 /* Ensure vcpu is not eligible during next spinloop */
1876 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001877}
1878EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1879
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001880static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001881{
1882 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001883 struct page *page;
1884
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001885 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001886 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001887#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001888 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001889 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001890#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001891#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1892 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1893 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1894#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001895 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001896 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001897 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001898 vmf->page = page;
1899 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001900}
1901
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001902static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001903 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001904};
1905
1906static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1907{
1908 vma->vm_ops = &kvm_vcpu_vm_ops;
1909 return 0;
1910}
1911
Avi Kivitybccf2152007-02-21 18:04:26 +02001912static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1913{
1914 struct kvm_vcpu *vcpu = filp->private_data;
1915
Al Viro66c0b392008-04-19 20:33:56 +01001916 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001917 return 0;
1918}
1919
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001920static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001921 .release = kvm_vcpu_release,
1922 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001923#ifdef CONFIG_COMPAT
1924 .compat_ioctl = kvm_vcpu_compat_ioctl,
1925#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001926 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001927 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001928};
1929
1930/*
1931 * Allocates an inode for the vcpu.
1932 */
1933static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1934{
Yann Droneaud24009b02013-08-24 22:14:07 +02001935 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001936}
1937
Avi Kivityc5ea7662007-02-20 18:41:05 +02001938/*
1939 * Creates some virtual cpus. Good luck creating more than one.
1940 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001941static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001942{
1943 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001944 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001945
Andy Honig338c7db2013-11-18 16:09:22 -08001946 if (id >= KVM_MAX_VCPUS)
1947 return -EINVAL;
1948
Gleb Natapov73880c82009-06-09 15:56:28 +03001949 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001950 if (IS_ERR(vcpu))
1951 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001952
Avi Kivity15ad7142007-07-11 18:17:21 +03001953 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1954
Avi Kivity26e52152007-11-20 15:30:24 +02001955 r = kvm_arch_vcpu_setup(vcpu);
1956 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001957 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001958
Shaohua Li11ec2802007-07-23 14:51:37 +08001959 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001960 if (!kvm_vcpu_compatible(vcpu)) {
1961 r = -EINVAL;
1962 goto unlock_vcpu_destroy;
1963 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001964 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1965 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001966 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001967 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001968
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001969 kvm_for_each_vcpu(r, v, kvm)
1970 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001971 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001972 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001973 }
1974
1975 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001976
1977 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001978 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001979 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001980 if (r < 0) {
1981 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001982 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001983 }
1984
1985 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1986 smp_wmb();
1987 atomic_inc(&kvm->online_vcpus);
1988
Gleb Natapov73880c82009-06-09 15:56:28 +03001989 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001990 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001991 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001992
Jan Kiszkad7805922011-05-23 10:33:05 +02001993unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001994 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001995vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001996 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001997 return r;
1998}
1999
Avi Kivity1961d272007-03-05 19:46:05 +02002000static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2001{
2002 if (sigset) {
2003 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2004 vcpu->sigset_active = 1;
2005 vcpu->sigset = *sigset;
2006 } else
2007 vcpu->sigset_active = 0;
2008 return 0;
2009}
2010
Avi Kivitybccf2152007-02-21 18:04:26 +02002011static long kvm_vcpu_ioctl(struct file *filp,
2012 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002013{
Avi Kivitybccf2152007-02-21 18:04:26 +02002014 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002015 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002016 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002017 struct kvm_fpu *fpu = NULL;
2018 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002019
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002020 if (vcpu->kvm->mm != current->mm)
2021 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002022
David Matlack2ea75be2014-09-19 16:03:25 -07002023 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2024 return -EINVAL;
2025
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002026#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002027 /*
2028 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2029 * so vcpu_load() would break it.
2030 */
2031 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2032 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2033#endif
2034
2035
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002036 r = vcpu_load(vcpu);
2037 if (r)
2038 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002039 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002040 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002041 r = -EINVAL;
2042 if (arg)
2043 goto out;
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002044 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2045 /* The thread running this VCPU changed. */
2046 struct pid *oldpid = vcpu->pid;
2047 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2048 rcu_assign_pointer(vcpu->pid, newpid);
2049 if (oldpid)
2050 synchronize_rcu();
2051 put_pid(oldpid);
2052 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002053 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002054 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002055 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002056 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002057 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002058
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002059 r = -ENOMEM;
2060 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2061 if (!kvm_regs)
2062 goto out;
2063 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002064 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002065 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002066 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002067 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2068 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002069 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002070out_free1:
2071 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002072 break;
2073 }
2074 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002075 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002076
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002077 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002078 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2079 if (IS_ERR(kvm_regs)) {
2080 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002081 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002082 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002083 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002084 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002085 break;
2086 }
2087 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002088 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2089 r = -ENOMEM;
2090 if (!kvm_sregs)
2091 goto out;
2092 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002093 if (r)
2094 goto out;
2095 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002096 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002097 goto out;
2098 r = 0;
2099 break;
2100 }
2101 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002102 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2103 if (IS_ERR(kvm_sregs)) {
2104 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002105 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002106 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002107 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002108 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002109 break;
2110 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002111 case KVM_GET_MP_STATE: {
2112 struct kvm_mp_state mp_state;
2113
2114 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2115 if (r)
2116 goto out;
2117 r = -EFAULT;
2118 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2119 goto out;
2120 r = 0;
2121 break;
2122 }
2123 case KVM_SET_MP_STATE: {
2124 struct kvm_mp_state mp_state;
2125
2126 r = -EFAULT;
2127 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2128 goto out;
2129 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002130 break;
2131 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002132 case KVM_TRANSLATE: {
2133 struct kvm_translation tr;
2134
2135 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002136 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002137 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002138 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002139 if (r)
2140 goto out;
2141 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002142 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002143 goto out;
2144 r = 0;
2145 break;
2146 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002147 case KVM_SET_GUEST_DEBUG: {
2148 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002149
2150 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002151 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002152 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002153 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002154 break;
2155 }
Avi Kivity1961d272007-03-05 19:46:05 +02002156 case KVM_SET_SIGNAL_MASK: {
2157 struct kvm_signal_mask __user *sigmask_arg = argp;
2158 struct kvm_signal_mask kvm_sigmask;
2159 sigset_t sigset, *p;
2160
2161 p = NULL;
2162 if (argp) {
2163 r = -EFAULT;
2164 if (copy_from_user(&kvm_sigmask, argp,
2165 sizeof kvm_sigmask))
2166 goto out;
2167 r = -EINVAL;
2168 if (kvm_sigmask.len != sizeof sigset)
2169 goto out;
2170 r = -EFAULT;
2171 if (copy_from_user(&sigset, sigmask_arg->sigset,
2172 sizeof sigset))
2173 goto out;
2174 p = &sigset;
2175 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002176 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002177 break;
2178 }
Avi Kivityb8836732007-04-01 16:34:31 +03002179 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002180 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2181 r = -ENOMEM;
2182 if (!fpu)
2183 goto out;
2184 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002185 if (r)
2186 goto out;
2187 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002188 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002189 goto out;
2190 r = 0;
2191 break;
2192 }
2193 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002194 fpu = memdup_user(argp, sizeof(*fpu));
2195 if (IS_ERR(fpu)) {
2196 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002197 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002198 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002199 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002200 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002201 break;
2202 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002203 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002204 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002205 }
2206out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002207 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002208 kfree(fpu);
2209 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002210 return r;
2211}
2212
Alexander Graf1dda6062011-06-08 02:45:37 +02002213#ifdef CONFIG_COMPAT
2214static long kvm_vcpu_compat_ioctl(struct file *filp,
2215 unsigned int ioctl, unsigned long arg)
2216{
2217 struct kvm_vcpu *vcpu = filp->private_data;
2218 void __user *argp = compat_ptr(arg);
2219 int r;
2220
2221 if (vcpu->kvm->mm != current->mm)
2222 return -EIO;
2223
2224 switch (ioctl) {
2225 case KVM_SET_SIGNAL_MASK: {
2226 struct kvm_signal_mask __user *sigmask_arg = argp;
2227 struct kvm_signal_mask kvm_sigmask;
2228 compat_sigset_t csigset;
2229 sigset_t sigset;
2230
2231 if (argp) {
2232 r = -EFAULT;
2233 if (copy_from_user(&kvm_sigmask, argp,
2234 sizeof kvm_sigmask))
2235 goto out;
2236 r = -EINVAL;
2237 if (kvm_sigmask.len != sizeof csigset)
2238 goto out;
2239 r = -EFAULT;
2240 if (copy_from_user(&csigset, sigmask_arg->sigset,
2241 sizeof csigset))
2242 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002243 sigset_from_compat(&sigset, &csigset);
2244 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2245 } else
2246 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002247 break;
2248 }
2249 default:
2250 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2251 }
2252
2253out:
2254 return r;
2255}
2256#endif
2257
Scott Wood852b6d52013-04-12 14:08:42 +00002258static int kvm_device_ioctl_attr(struct kvm_device *dev,
2259 int (*accessor)(struct kvm_device *dev,
2260 struct kvm_device_attr *attr),
2261 unsigned long arg)
2262{
2263 struct kvm_device_attr attr;
2264
2265 if (!accessor)
2266 return -EPERM;
2267
2268 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2269 return -EFAULT;
2270
2271 return accessor(dev, &attr);
2272}
2273
2274static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2275 unsigned long arg)
2276{
2277 struct kvm_device *dev = filp->private_data;
2278
2279 switch (ioctl) {
2280 case KVM_SET_DEVICE_ATTR:
2281 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2282 case KVM_GET_DEVICE_ATTR:
2283 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2284 case KVM_HAS_DEVICE_ATTR:
2285 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2286 default:
2287 if (dev->ops->ioctl)
2288 return dev->ops->ioctl(dev, ioctl, arg);
2289
2290 return -ENOTTY;
2291 }
2292}
2293
Scott Wood852b6d52013-04-12 14:08:42 +00002294static int kvm_device_release(struct inode *inode, struct file *filp)
2295{
2296 struct kvm_device *dev = filp->private_data;
2297 struct kvm *kvm = dev->kvm;
2298
Scott Wood852b6d52013-04-12 14:08:42 +00002299 kvm_put_kvm(kvm);
2300 return 0;
2301}
2302
2303static const struct file_operations kvm_device_fops = {
2304 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002305#ifdef CONFIG_COMPAT
2306 .compat_ioctl = kvm_device_ioctl,
2307#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002308 .release = kvm_device_release,
2309};
2310
2311struct kvm_device *kvm_device_from_filp(struct file *filp)
2312{
2313 if (filp->f_op != &kvm_device_fops)
2314 return NULL;
2315
2316 return filp->private_data;
2317}
2318
Will Deacond60eacb2014-09-02 10:27:33 +01002319static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2320#ifdef CONFIG_KVM_MPIC
2321 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2322 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2323#endif
2324
2325#ifdef CONFIG_KVM_XICS
2326 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2327#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002328};
2329
2330int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2331{
2332 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2333 return -ENOSPC;
2334
2335 if (kvm_device_ops_table[type] != NULL)
2336 return -EEXIST;
2337
2338 kvm_device_ops_table[type] = ops;
2339 return 0;
2340}
2341
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002342void kvm_unregister_device_ops(u32 type)
2343{
2344 if (kvm_device_ops_table[type] != NULL)
2345 kvm_device_ops_table[type] = NULL;
2346}
2347
Scott Wood852b6d52013-04-12 14:08:42 +00002348static int kvm_ioctl_create_device(struct kvm *kvm,
2349 struct kvm_create_device *cd)
2350{
2351 struct kvm_device_ops *ops = NULL;
2352 struct kvm_device *dev;
2353 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2354 int ret;
2355
Will Deacond60eacb2014-09-02 10:27:33 +01002356 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002357 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002358
2359 ops = kvm_device_ops_table[cd->type];
2360 if (ops == NULL)
2361 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002362
2363 if (test)
2364 return 0;
2365
2366 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2367 if (!dev)
2368 return -ENOMEM;
2369
2370 dev->ops = ops;
2371 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002372
2373 ret = ops->create(dev, cd->type);
2374 if (ret < 0) {
2375 kfree(dev);
2376 return ret;
2377 }
2378
Yann Droneaud24009b02013-08-24 22:14:07 +02002379 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002380 if (ret < 0) {
2381 ops->destroy(dev);
2382 return ret;
2383 }
2384
Scott Wood07f0a7b2013-04-25 14:11:23 +00002385 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002386 kvm_get_kvm(kvm);
2387 cd->fd = ret;
2388 return 0;
2389}
2390
Alexander Graf92b591a2014-07-14 18:33:08 +02002391static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2392{
2393 switch (arg) {
2394 case KVM_CAP_USER_MEMORY:
2395 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2396 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2397#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2398 case KVM_CAP_SET_BOOT_CPU_ID:
2399#endif
2400 case KVM_CAP_INTERNAL_ERROR_DATA:
2401#ifdef CONFIG_HAVE_KVM_MSI
2402 case KVM_CAP_SIGNAL_MSI:
2403#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002404#ifdef CONFIG_HAVE_KVM_IRQFD
Alexander Graf92b591a2014-07-14 18:33:08 +02002405 case KVM_CAP_IRQFD_RESAMPLE:
2406#endif
2407 case KVM_CAP_CHECK_EXTENSION_VM:
2408 return 1;
2409#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2410 case KVM_CAP_IRQ_ROUTING:
2411 return KVM_MAX_IRQ_ROUTES;
2412#endif
2413 default:
2414 break;
2415 }
2416 return kvm_vm_ioctl_check_extension(kvm, arg);
2417}
2418
Avi Kivitybccf2152007-02-21 18:04:26 +02002419static long kvm_vm_ioctl(struct file *filp,
2420 unsigned int ioctl, unsigned long arg)
2421{
2422 struct kvm *kvm = filp->private_data;
2423 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002424 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002425
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002426 if (kvm->mm != current->mm)
2427 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002428 switch (ioctl) {
2429 case KVM_CREATE_VCPU:
2430 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002431 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002432 case KVM_SET_USER_MEMORY_REGION: {
2433 struct kvm_userspace_memory_region kvm_userspace_mem;
2434
2435 r = -EFAULT;
2436 if (copy_from_user(&kvm_userspace_mem, argp,
2437 sizeof kvm_userspace_mem))
2438 goto out;
2439
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002440 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002441 break;
2442 }
2443 case KVM_GET_DIRTY_LOG: {
2444 struct kvm_dirty_log log;
2445
2446 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002447 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002448 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002449 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002450 break;
2451 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002452#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2453 case KVM_REGISTER_COALESCED_MMIO: {
2454 struct kvm_coalesced_mmio_zone zone;
2455 r = -EFAULT;
2456 if (copy_from_user(&zone, argp, sizeof zone))
2457 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002458 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002459 break;
2460 }
2461 case KVM_UNREGISTER_COALESCED_MMIO: {
2462 struct kvm_coalesced_mmio_zone zone;
2463 r = -EFAULT;
2464 if (copy_from_user(&zone, argp, sizeof zone))
2465 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002466 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002467 break;
2468 }
2469#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002470 case KVM_IRQFD: {
2471 struct kvm_irqfd data;
2472
2473 r = -EFAULT;
2474 if (copy_from_user(&data, argp, sizeof data))
2475 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002476 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002477 break;
2478 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002479 case KVM_IOEVENTFD: {
2480 struct kvm_ioeventfd data;
2481
2482 r = -EFAULT;
2483 if (copy_from_user(&data, argp, sizeof data))
2484 goto out;
2485 r = kvm_ioeventfd(kvm, &data);
2486 break;
2487 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002488#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2489 case KVM_SET_BOOT_CPU_ID:
2490 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002491 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002492 if (atomic_read(&kvm->online_vcpus) != 0)
2493 r = -EBUSY;
2494 else
2495 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002496 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002497 break;
2498#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002499#ifdef CONFIG_HAVE_KVM_MSI
2500 case KVM_SIGNAL_MSI: {
2501 struct kvm_msi msi;
2502
2503 r = -EFAULT;
2504 if (copy_from_user(&msi, argp, sizeof msi))
2505 goto out;
2506 r = kvm_send_userspace_msi(kvm, &msi);
2507 break;
2508 }
2509#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002510#ifdef __KVM_HAVE_IRQ_LINE
2511 case KVM_IRQ_LINE_STATUS:
2512 case KVM_IRQ_LINE: {
2513 struct kvm_irq_level irq_event;
2514
2515 r = -EFAULT;
2516 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2517 goto out;
2518
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002519 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2520 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002521 if (r)
2522 goto out;
2523
2524 r = -EFAULT;
2525 if (ioctl == KVM_IRQ_LINE_STATUS) {
2526 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2527 goto out;
2528 }
2529
2530 r = 0;
2531 break;
2532 }
2533#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002534#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2535 case KVM_SET_GSI_ROUTING: {
2536 struct kvm_irq_routing routing;
2537 struct kvm_irq_routing __user *urouting;
2538 struct kvm_irq_routing_entry *entries;
2539
2540 r = -EFAULT;
2541 if (copy_from_user(&routing, argp, sizeof(routing)))
2542 goto out;
2543 r = -EINVAL;
2544 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2545 goto out;
2546 if (routing.flags)
2547 goto out;
2548 r = -ENOMEM;
2549 entries = vmalloc(routing.nr * sizeof(*entries));
2550 if (!entries)
2551 goto out;
2552 r = -EFAULT;
2553 urouting = argp;
2554 if (copy_from_user(entries, urouting->entries,
2555 routing.nr * sizeof(*entries)))
2556 goto out_free_irq_routing;
2557 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2558 routing.flags);
2559 out_free_irq_routing:
2560 vfree(entries);
2561 break;
2562 }
2563#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002564 case KVM_CREATE_DEVICE: {
2565 struct kvm_create_device cd;
2566
2567 r = -EFAULT;
2568 if (copy_from_user(&cd, argp, sizeof(cd)))
2569 goto out;
2570
2571 r = kvm_ioctl_create_device(kvm, &cd);
2572 if (r)
2573 goto out;
2574
2575 r = -EFAULT;
2576 if (copy_to_user(argp, &cd, sizeof(cd)))
2577 goto out;
2578
2579 r = 0;
2580 break;
2581 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002582 case KVM_CHECK_EXTENSION:
2583 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2584 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002585 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002586 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002587 }
2588out:
2589 return r;
2590}
2591
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002592#ifdef CONFIG_COMPAT
2593struct compat_kvm_dirty_log {
2594 __u32 slot;
2595 __u32 padding1;
2596 union {
2597 compat_uptr_t dirty_bitmap; /* one bit per page */
2598 __u64 padding2;
2599 };
2600};
2601
2602static long kvm_vm_compat_ioctl(struct file *filp,
2603 unsigned int ioctl, unsigned long arg)
2604{
2605 struct kvm *kvm = filp->private_data;
2606 int r;
2607
2608 if (kvm->mm != current->mm)
2609 return -EIO;
2610 switch (ioctl) {
2611 case KVM_GET_DIRTY_LOG: {
2612 struct compat_kvm_dirty_log compat_log;
2613 struct kvm_dirty_log log;
2614
2615 r = -EFAULT;
2616 if (copy_from_user(&compat_log, (void __user *)arg,
2617 sizeof(compat_log)))
2618 goto out;
2619 log.slot = compat_log.slot;
2620 log.padding1 = compat_log.padding1;
2621 log.padding2 = compat_log.padding2;
2622 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2623
2624 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002625 break;
2626 }
2627 default:
2628 r = kvm_vm_ioctl(filp, ioctl, arg);
2629 }
2630
2631out:
2632 return r;
2633}
2634#endif
2635
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002636static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002637 .release = kvm_vm_release,
2638 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002639#ifdef CONFIG_COMPAT
2640 .compat_ioctl = kvm_vm_compat_ioctl,
2641#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002642 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002643};
2644
Carsten Ottee08b9632012-01-04 10:25:20 +01002645static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002646{
Heiko Carstensaac87632010-10-27 17:22:10 +02002647 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002648 struct kvm *kvm;
2649
Carsten Ottee08b9632012-01-04 10:25:20 +01002650 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002651 if (IS_ERR(kvm))
2652 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002653#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2654 r = kvm_coalesced_mmio_init(kvm);
2655 if (r < 0) {
2656 kvm_put_kvm(kvm);
2657 return r;
2658 }
2659#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002660 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002661 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002662 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002663
Heiko Carstensaac87632010-10-27 17:22:10 +02002664 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002665}
2666
2667static long kvm_dev_ioctl(struct file *filp,
2668 unsigned int ioctl, unsigned long arg)
2669{
Avi Kivity07c45a32007-03-07 13:05:38 +02002670 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002671
2672 switch (ioctl) {
2673 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002674 if (arg)
2675 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002676 r = KVM_API_VERSION;
2677 break;
2678 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002679 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002680 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002681 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02002682 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002683 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002684 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02002685 if (arg)
2686 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002687 r = PAGE_SIZE; /* struct kvm_run */
2688#ifdef CONFIG_X86
2689 r += PAGE_SIZE; /* pio data page */
2690#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002691#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2692 r += PAGE_SIZE; /* coalesced mmio ring page */
2693#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002694 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002695 case KVM_TRACE_ENABLE:
2696 case KVM_TRACE_PAUSE:
2697 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002698 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002699 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002700 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002701 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002702 }
2703out:
2704 return r;
2705}
2706
Avi Kivity6aa8b732006-12-10 02:21:36 -08002707static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002708 .unlocked_ioctl = kvm_dev_ioctl,
2709 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002710 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002711};
2712
2713static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002714 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002715 "kvm",
2716 &kvm_chardev_ops,
2717};
2718
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002719static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002720{
2721 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002722 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002723
Rusty Russell7f59f492008-12-07 21:25:45 +10302724 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002725 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002726
Rusty Russell7f59f492008-12-07 21:25:45 +10302727 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002728
Radim Krčmář13a34e02014-08-28 15:13:03 +02002729 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02002730
2731 if (r) {
2732 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2733 atomic_inc(&hardware_enable_failed);
2734 printk(KERN_INFO "kvm: enabling virtualization on "
2735 "CPU%d failed\n", cpu);
2736 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002737}
2738
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002739static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002740{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002741 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002742 if (kvm_usage_count)
2743 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002744 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002745}
2746
2747static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002748{
2749 int cpu = raw_smp_processor_id();
2750
Rusty Russell7f59f492008-12-07 21:25:45 +10302751 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002752 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302753 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02002754 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03002755}
2756
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002757static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002758{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002759 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002760 if (kvm_usage_count)
2761 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002762 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002763}
2764
Alexander Graf10474ae2009-09-15 11:37:46 +02002765static void hardware_disable_all_nolock(void)
2766{
2767 BUG_ON(!kvm_usage_count);
2768
2769 kvm_usage_count--;
2770 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002771 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002772}
2773
2774static void hardware_disable_all(void)
2775{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002776 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002777 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002778 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002779}
2780
2781static int hardware_enable_all(void)
2782{
2783 int r = 0;
2784
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002785 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002786
2787 kvm_usage_count++;
2788 if (kvm_usage_count == 1) {
2789 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002790 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002791
2792 if (atomic_read(&hardware_enable_failed)) {
2793 hardware_disable_all_nolock();
2794 r = -EBUSY;
2795 }
2796 }
2797
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002798 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002799
2800 return r;
2801}
2802
Avi Kivity774c47f2007-02-12 00:54:47 -08002803static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2804 void *v)
2805{
2806 int cpu = (long)v;
2807
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002808 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002809 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002810 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002811 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2812 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002813 hardware_disable();
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002814 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002815 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002816 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2817 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002818 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002819 break;
2820 }
2821 return NOTIFY_OK;
2822}
2823
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002824static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002825 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002826{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002827 /*
2828 * Some (well, at least mine) BIOSes hang on reboot if
2829 * in vmx root mode.
2830 *
2831 * And Intel TXT required VMX off for all cpu when system shutdown.
2832 */
2833 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2834 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002835 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002836 return NOTIFY_OK;
2837}
2838
2839static struct notifier_block kvm_reboot_notifier = {
2840 .notifier_call = kvm_reboot,
2841 .priority = 0,
2842};
2843
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002844static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002845{
2846 int i;
2847
2848 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002849 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002850
2851 kvm_iodevice_destructor(pos);
2852 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002853 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002854}
2855
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002856static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2857 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002858{
Sasha Levin743eeb02011-07-27 16:00:48 +03002859 if (r1->addr < r2->addr)
2860 return -1;
2861 if (r1->addr + r1->len > r2->addr + r2->len)
2862 return 1;
2863 return 0;
2864}
2865
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002866static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2867{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002868 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002869}
2870
Geoff Levand39369f72013-04-05 19:20:30 +00002871static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002872 gpa_t addr, int len)
2873{
Sasha Levin743eeb02011-07-27 16:00:48 +03002874 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2875 .addr = addr,
2876 .len = len,
2877 .dev = dev,
2878 };
2879
2880 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2881 kvm_io_bus_sort_cmp, NULL);
2882
2883 return 0;
2884}
2885
Geoff Levand39369f72013-04-05 19:20:30 +00002886static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002887 gpa_t addr, int len)
2888{
2889 struct kvm_io_range *range, key;
2890 int off;
2891
2892 key = (struct kvm_io_range) {
2893 .addr = addr,
2894 .len = len,
2895 };
2896
2897 range = bsearch(&key, bus->range, bus->dev_count,
2898 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2899 if (range == NULL)
2900 return -ENOENT;
2901
2902 off = range - bus->range;
2903
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002904 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002905 off--;
2906
2907 return off;
2908}
2909
Cornelia Huck126a5af2013-07-03 16:30:53 +02002910static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2911 struct kvm_io_range *range, const void *val)
2912{
2913 int idx;
2914
2915 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2916 if (idx < 0)
2917 return -EOPNOTSUPP;
2918
2919 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002920 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002921 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2922 range->len, val))
2923 return idx;
2924 idx++;
2925 }
2926
2927 return -EOPNOTSUPP;
2928}
2929
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002930/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002931int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002932 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002933{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002934 struct kvm_io_bus *bus;
2935 struct kvm_io_range range;
2936 int r;
2937
2938 range = (struct kvm_io_range) {
2939 .addr = addr,
2940 .len = len,
2941 };
2942
2943 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2944 r = __kvm_io_bus_write(bus, &range, val);
2945 return r < 0 ? r : 0;
2946}
2947
2948/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2949int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2950 int len, const void *val, long cookie)
2951{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002952 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002953 struct kvm_io_range range;
2954
2955 range = (struct kvm_io_range) {
2956 .addr = addr,
2957 .len = len,
2958 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002959
2960 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002961
2962 /* First try the device referenced by cookie. */
2963 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002964 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002965 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2966 val))
2967 return cookie;
2968
2969 /*
2970 * cookie contained garbage; fall back to search and return the
2971 * correct cookie value.
2972 */
2973 return __kvm_io_bus_write(bus, &range, val);
2974}
2975
2976static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2977 void *val)
2978{
2979 int idx;
2980
2981 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002982 if (idx < 0)
2983 return -EOPNOTSUPP;
2984
2985 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002986 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002987 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2988 range->len, val))
2989 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03002990 idx++;
2991 }
2992
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002993 return -EOPNOTSUPP;
2994}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03002995EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002996
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002997/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002998int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2999 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003000{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003001 struct kvm_io_bus *bus;
3002 struct kvm_io_range range;
3003 int r;
3004
3005 range = (struct kvm_io_range) {
3006 .addr = addr,
3007 .len = len,
3008 };
3009
3010 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3011 r = __kvm_io_bus_read(bus, &range, val);
3012 return r < 0 ? r : 0;
3013}
3014
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003015
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003016/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003017int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3018 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003019{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003020 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003021
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003022 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08003023 /* exclude ioeventfd which is limited by maximum fd */
3024 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003025 return -ENOSPC;
3026
Amos Konga13007162012-03-09 12:17:32 +08003027 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3028 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003029 if (!new_bus)
3030 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003031 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3032 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003033 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003034 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3035 synchronize_srcu_expedited(&kvm->srcu);
3036 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003037
3038 return 0;
3039}
3040
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003041/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003042int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3043 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003044{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003045 int i, r;
3046 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003047
Sasha Levincdfca7b2011-12-04 19:36:28 +02003048 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003049 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003050 for (i = 0; i < bus->dev_count; i++)
3051 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003052 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003053 break;
3054 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003055
Amos Konga13007162012-03-09 12:17:32 +08003056 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003057 return r;
Amos Konga13007162012-03-09 12:17:32 +08003058
3059 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3060 sizeof(struct kvm_io_range)), GFP_KERNEL);
3061 if (!new_bus)
3062 return -ENOMEM;
3063
3064 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3065 new_bus->dev_count--;
3066 memcpy(new_bus->range + i, bus->range + i + 1,
3067 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003068
3069 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3070 synchronize_srcu_expedited(&kvm->srcu);
3071 kfree(bus);
3072 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003073}
3074
Avi Kivity774c47f2007-02-12 00:54:47 -08003075static struct notifier_block kvm_cpu_notifier = {
3076 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003077};
3078
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003079static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003080{
3081 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003082 struct kvm *kvm;
3083
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003084 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003085 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003086 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003087 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003088 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003089 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003090}
3091
3092DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3093
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003094static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003095{
3096 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003097 struct kvm *kvm;
3098 struct kvm_vcpu *vcpu;
3099 int i;
3100
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003101 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003102 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003103 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003104 kvm_for_each_vcpu(i, vcpu, kvm)
3105 *val += *(u32 *)((void *)vcpu + offset);
3106
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003107 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003108 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003109}
3110
Avi Kivityba1389b2007-11-18 16:24:12 +02003111DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3112
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003113static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003114 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3115 [KVM_STAT_VM] = &vm_stat_fops,
3116};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003117
Hamo4f69b682011-12-15 14:23:16 +08003118static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003119{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003120 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003121 struct kvm_stats_debugfs_item *p;
3122
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003123 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003124 if (kvm_debugfs_dir == NULL)
3125 goto out;
3126
3127 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003128 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003129 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003130 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003131 if (p->dentry == NULL)
3132 goto out_dir;
3133 }
3134
3135 return 0;
3136
3137out_dir:
3138 debugfs_remove_recursive(kvm_debugfs_dir);
3139out:
3140 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003141}
3142
3143static void kvm_exit_debug(void)
3144{
3145 struct kvm_stats_debugfs_item *p;
3146
3147 for (p = debugfs_entries; p->name; ++p)
3148 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003149 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003150}
3151
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003152static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003153{
Alexander Graf10474ae2009-09-15 11:37:46 +02003154 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003155 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003156 return 0;
3157}
3158
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003159static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003160{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003161 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003162 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003163 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003164 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003165}
3166
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003167static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003168 .suspend = kvm_suspend,
3169 .resume = kvm_resume,
3170};
3171
Avi Kivity15ad7142007-07-11 18:17:21 +03003172static inline
3173struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3174{
3175 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3176}
3177
3178static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3179{
3180 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303181 if (vcpu->preempted)
3182 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003183
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003184 kvm_arch_sched_in(vcpu, cpu);
3185
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003186 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003187}
3188
3189static void kvm_sched_out(struct preempt_notifier *pn,
3190 struct task_struct *next)
3191{
3192 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3193
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303194 if (current->state == TASK_RUNNING)
3195 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003196 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003197}
3198
Avi Kivity0ee75be2010-04-28 15:39:01 +03003199int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003200 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003201{
3202 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003203 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003204
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003205 r = kvm_arch_init(opaque);
3206 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003207 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003208
Asias He7dac16c2013-05-08 10:57:29 +08003209 /*
3210 * kvm_arch_init makes sure there's at most one caller
3211 * for architectures that support multiple implementations,
3212 * like intel and amd on x86.
3213 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3214 * conflicts in case kvm is already setup for another implementation.
3215 */
3216 r = kvm_irqfd_init();
3217 if (r)
3218 goto out_irqfd;
3219
Avi Kivity8437a612009-06-06 14:52:35 -07003220 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303221 r = -ENOMEM;
3222 goto out_free_0;
3223 }
3224
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003225 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003226 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303227 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003228
Yang, Sheng002c7f72007-07-31 14:23:01 +03003229 for_each_online_cpu(cpu) {
3230 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003231 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003232 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003233 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003234 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003235 }
3236
Avi Kivity774c47f2007-02-12 00:54:47 -08003237 r = register_cpu_notifier(&kvm_cpu_notifier);
3238 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003239 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003240 register_reboot_notifier(&kvm_reboot_notifier);
3241
Rusty Russellc16f8622007-07-30 21:12:19 +10003242 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003243 if (!vcpu_align)
3244 vcpu_align = __alignof__(struct kvm_vcpu);
3245 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003246 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003247 if (!kvm_vcpu_cache) {
3248 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003249 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003250 }
3251
Gleb Natapovaf585b92010-10-14 11:22:46 +02003252 r = kvm_async_pf_init();
3253 if (r)
3254 goto out_free;
3255
Avi Kivity6aa8b732006-12-10 02:21:36 -08003256 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003257 kvm_vm_fops.owner = module;
3258 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003259
3260 r = misc_register(&kvm_dev);
3261 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003262 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003263 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003264 }
3265
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003266 register_syscore_ops(&kvm_syscore_ops);
3267
Avi Kivity15ad7142007-07-11 18:17:21 +03003268 kvm_preempt_ops.sched_in = kvm_sched_in;
3269 kvm_preempt_ops.sched_out = kvm_sched_out;
3270
Hamo4f69b682011-12-15 14:23:16 +08003271 r = kvm_init_debug();
3272 if (r) {
3273 printk(KERN_ERR "kvm: create debugfs files failed\n");
3274 goto out_undebugfs;
3275 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003276
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003277 r = kvm_vfio_ops_init();
3278 WARN_ON(r);
3279
Avi Kivityc7addb92007-09-16 18:58:32 +02003280 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003281
Hamo4f69b682011-12-15 14:23:16 +08003282out_undebugfs:
3283 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003284 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003285out_unreg:
3286 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003287out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003288 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003289out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003290 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003291 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003292out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003293out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003294 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303295out_free_0a:
3296 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003297out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003298 kvm_irqfd_exit();
3299out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003300 kvm_arch_exit();
3301out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003302 return r;
3303}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003304EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003305
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003306void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003307{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003308 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003309 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003310 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003311 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003312 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003313 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003314 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003315 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003316 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003317 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003318 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303319 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08003320 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003321}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003322EXPORT_SYMBOL_GPL(kvm_exit);