blob: 25ffac9e947d9d3e2d554e6c351dfa51811c0354 [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
Xiao Guangronga2766322012-07-26 11:58:59 +0800110bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300111{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200112 if (pfn_valid(pfn))
Ard Biesheuvel85c85552014-09-12 15:16:00 +0200113 return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300114
115 return true;
116}
117
Avi Kivity6aa8b732006-12-10 02:21:36 -0800118/*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300121int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800122{
Avi Kivity15ad7142007-07-11 18:17:21 +0300123 int cpu;
124
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
Christian Borntraeger7103f602014-08-19 16:45:56 +0200132 if (oldpid)
133 synchronize_rcu();
Rik van Riel34bb10b2011-02-01 09:52:41 -0500134 put_pid(oldpid);
135 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 cpu = get_cpu();
137 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200138 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300139 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300140 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200141}
142
Carsten Otte313a3dc2007-10-11 19:16:52 +0200143void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800144{
Avi Kivity15ad7142007-07-11 18:17:21 +0300145 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200146 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300147 preempt_notifier_unregister(&vcpu->preempt_notifier);
148 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800149 mutex_unlock(&vcpu->mutex);
150}
151
Avi Kivityd9e368d2007-06-07 19:18:30 +0300152static void ack_flush(void *_completed)
153{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300154}
155
Tang Chen445b8232014-09-24 15:57:55 +0800156bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300157{
Avi Kivity597a5f52008-07-20 14:24:22 +0300158 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030159 cpumask_var_t cpus;
160 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300161 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300162
Li Zefan79f55992009-06-15 14:58:26 +0800163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030164
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800165 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300166 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800167 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300168 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800169
170 /* Set ->requests bit before we read ->mode */
171 smp_mb();
172
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030175 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300176 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
181 else
182 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800183 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030184 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030185 return called;
186}
187
188void kvm_flush_remote_tlbs(struct kvm *kvm)
189{
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800190 long dirty_count = kvm->tlbs_dirty;
191
192 smp_mb();
Tang Chen445b8232014-09-24 15:57:55 +0800193 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
Rusty Russell49846892008-12-08 20:26:24 +1030194 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300196}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530197EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300198
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500199void kvm_reload_remote_mmus(struct kvm *kvm)
200{
Tang Chen445b8232014-09-24 15:57:55 +0800201 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500202}
203
Marcelo Tosattid8281992012-11-27 23:29:01 -0200204void kvm_make_mclock_inprogress_request(struct kvm *kvm)
205{
Tang Chen445b8232014-09-24 15:57:55 +0800206 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
Marcelo Tosattid8281992012-11-27 23:29:01 -0200207}
208
Yang Zhang3d81bc72013-04-11 19:25:13 +0800209void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800210{
Tang Chen445b8232014-09-24 15:57:55 +0800211 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800212}
213
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000214int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215{
216 struct page *page;
217 int r;
218
219 mutex_init(&vcpu->mutex);
220 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000221 vcpu->kvm = kvm;
222 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500223 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300224 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200225 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000226
227 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 if (!page) {
229 r = -ENOMEM;
230 goto fail;
231 }
232 vcpu->run = page_address(page);
233
Raghavendra K T4c088492012-07-18 19:07:46 +0530234 kvm_vcpu_set_in_spin_loop(vcpu, false);
235 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530236 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530237
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800238 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000239 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800240 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000241 return 0;
242
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000243fail_free_run:
244 free_page((unsigned long)vcpu->run);
245fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000246 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000247}
248EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500252 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800253 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000254 free_page((unsigned long)vcpu->run);
255}
256EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200258#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260{
261 return container_of(mn, struct kvm, mmu_notifier);
262}
263
264static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
267{
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200269 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200270
271 /*
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * before returning.
280 *
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
283 *
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
288 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200289 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200290 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900291
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200292 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200294 /* we've to flush the tlb before the pages can be freed */
295 if (need_tlb_flush)
296 kvm_flush_remote_tlbs(kvm);
297
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900298 spin_unlock(&kvm->mmu_lock);
Tang Chenfe715572014-09-24 15:57:57 +0800299
300 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
301
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900302 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200303}
304
Izik Eidus3da0dd42009-09-23 21:47:18 +0300305static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
306 struct mm_struct *mm,
307 unsigned long address,
308 pte_t pte)
309{
310 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200311 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300312
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200313 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300314 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 kvm_set_spte_hva(kvm, address, pte);
317 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200318 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300319}
320
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200321static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
322 struct mm_struct *mm,
323 unsigned long start,
324 unsigned long end)
325{
326 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200327 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200328
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200329 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200330 spin_lock(&kvm->mmu_lock);
331 /*
332 * The count increase must become visible at unlock time as no
333 * spte can be established without taking the mmu_lock and
334 * count is also read inside the mmu_lock critical section.
335 */
336 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900337 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800338 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200339 /* we've to flush the tlb before the pages can be freed */
340 if (need_tlb_flush)
341 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900342
343 spin_unlock(&kvm->mmu_lock);
344 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200345}
346
347static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
348 struct mm_struct *mm,
349 unsigned long start,
350 unsigned long end)
351{
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
353
354 spin_lock(&kvm->mmu_lock);
355 /*
356 * This sequence increase will notify the kvm page fault that
357 * the page that is going to be mapped in the spte could have
358 * been freed.
359 */
360 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000361 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200362 /*
363 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000364 * below count decrease, which is ensured by the smp_wmb above
365 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200366 */
367 kvm->mmu_notifier_count--;
368 spin_unlock(&kvm->mmu_lock);
369
370 BUG_ON(kvm->mmu_notifier_count < 0);
371}
372
373static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
374 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700375 unsigned long start,
376 unsigned long end)
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200377{
378 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200379 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200380
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200381 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200382 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200383
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700384 young = kvm_age_hva(kvm, start, end);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200385 if (young)
386 kvm_flush_remote_tlbs(kvm);
387
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900388 spin_unlock(&kvm->mmu_lock);
389 srcu_read_unlock(&kvm->srcu, idx);
390
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200391 return young;
392}
393
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800394static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
395 struct mm_struct *mm,
396 unsigned long address)
397{
398 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 int young, idx;
400
401 idx = srcu_read_lock(&kvm->srcu);
402 spin_lock(&kvm->mmu_lock);
403 young = kvm_test_age_hva(kvm, address);
404 spin_unlock(&kvm->mmu_lock);
405 srcu_read_unlock(&kvm->srcu, idx);
406
407 return young;
408}
409
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100410static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
411 struct mm_struct *mm)
412{
413 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800414 int idx;
415
416 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300417 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800418 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100419}
420
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200421static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
422 .invalidate_page = kvm_mmu_notifier_invalidate_page,
423 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
424 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
425 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800426 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300427 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100428 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200429};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200430
431static int kvm_init_mmu_notifier(struct kvm *kvm)
432{
433 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
434 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
435}
436
437#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
438
439static int kvm_init_mmu_notifier(struct kvm *kvm)
440{
441 return 0;
442}
443
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200444#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
445
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800446static void kvm_init_memslots_id(struct kvm *kvm)
447{
448 int i;
449 struct kvm_memslots *slots = kvm->memslots;
450
451 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800452 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800453}
454
Carsten Ottee08b9632012-01-04 10:25:20 +0100455static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800456{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100457 int r, i;
458 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800459
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100460 if (!kvm)
461 return ERR_PTR(-ENOMEM);
462
Carsten Ottee08b9632012-01-04 10:25:20 +0100463 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100464 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100465 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200466
467 r = hardware_enable_all();
468 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100469 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200470
Avi Kivity75858a82009-01-04 17:10:50 +0200471#ifdef CONFIG_HAVE_KVM_IRQCHIP
472 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200473#endif
474#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300475 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200476#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800477
Alex Williamson1e702d92012-12-10 10:33:32 -0700478 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
479
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200480 r = -ENOMEM;
481 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
482 if (!kvm->memslots)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100483 goto out_err_no_srcu;
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200484
485 /*
486 * Init kvm generation close to the maximum to easily test the
487 * code of handling generation number wrap-around.
488 */
489 kvm->memslots->generation = -150;
490
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800491 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200492 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100493 goto out_err_no_srcu;
494 if (init_srcu_struct(&kvm->irq_srcu))
495 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200496 for (i = 0; i < KVM_NR_BUSES; i++) {
497 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
498 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100499 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200500 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200501 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700502
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900503 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200504 kvm->mm = current->mm;
505 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400506 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800507 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300508 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200509 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300510 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000511 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700512
513 r = kvm_init_mmu_notifier(kvm);
514 if (r)
515 goto out_err;
516
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200517 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000518 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200519 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100520
Avi Kivityf17abe92007-02-21 19:28:04 +0200521 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200522
523out_err:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100524 cleanup_srcu_struct(&kvm->irq_srcu);
525out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100526 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100527out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200528 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100529out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200530 for (i = 0; i < KVM_NR_BUSES; i++)
531 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200532 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100533 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200534 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200535}
536
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900537/*
538 * Avoid using vmalloc for a small buffer.
539 * Should not be used when the size is statically known.
540 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900541void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900542{
543 if (size > PAGE_SIZE)
544 return vzalloc(size);
545 else
546 return kzalloc(size, GFP_KERNEL);
547}
548
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900549void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900550{
551 if (is_vmalloc_addr(addr))
552 vfree(addr);
553 else
554 kfree(addr);
555}
556
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900557static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
558{
559 if (!memslot->dirty_bitmap)
560 return;
561
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900562 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900563 memslot->dirty_bitmap = NULL;
564}
565
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566/*
567 * Free any memory in @free but not in @dont.
568 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530569static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800570 struct kvm_memory_slot *dont)
571{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800572 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900573 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800574
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530575 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300576
Avi Kivity6aa8b732006-12-10 02:21:36 -0800577 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800578}
579
Stephen Hemminger79408762013-12-29 12:12:29 -0800580static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800581{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200582 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800583 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800584
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800585 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530586 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200587
588 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800589}
590
Scott Wood07f0a7b2013-04-25 14:11:23 +0000591static void kvm_destroy_devices(struct kvm *kvm)
592{
593 struct list_head *node, *tmp;
594
595 list_for_each_safe(node, tmp, &kvm->devices) {
596 struct kvm_device *dev =
597 list_entry(node, struct kvm_device, vm_node);
598
599 list_del(node);
600 dev->ops->destroy(dev);
601 }
602}
603
Avi Kivityf17abe92007-02-21 19:28:04 +0200604static void kvm_destroy_vm(struct kvm *kvm)
605{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200606 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200607 struct mm_struct *mm = kvm->mm;
608
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800609 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200610 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800611 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200612 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200613 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200614 for (i = 0; i < KVM_NR_BUSES; i++)
615 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200616 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200617#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
618 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200619#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300620 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200621#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800622 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000623 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100624 kvm_free_physmem(kvm);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200625 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100626 cleanup_srcu_struct(&kvm->srcu);
627 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200628 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200629 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200630}
631
Izik Eidusd39f13b2008-03-30 16:01:25 +0300632void kvm_get_kvm(struct kvm *kvm)
633{
634 atomic_inc(&kvm->users_count);
635}
636EXPORT_SYMBOL_GPL(kvm_get_kvm);
637
638void kvm_put_kvm(struct kvm *kvm)
639{
640 if (atomic_dec_and_test(&kvm->users_count))
641 kvm_destroy_vm(kvm);
642}
643EXPORT_SYMBOL_GPL(kvm_put_kvm);
644
645
Avi Kivityf17abe92007-02-21 19:28:04 +0200646static int kvm_vm_release(struct inode *inode, struct file *filp)
647{
648 struct kvm *kvm = filp->private_data;
649
Gregory Haskins721eecb2009-05-20 10:30:49 -0400650 kvm_irqfd_release(kvm);
651
Izik Eidusd39f13b2008-03-30 16:01:25 +0300652 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800653 return 0;
654}
655
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900656/*
657 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900658 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900659 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900660static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
661{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900662 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900663
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900664 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900665 if (!memslot->dirty_bitmap)
666 return -ENOMEM;
667
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900668 return 0;
669}
670
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800671static int cmp_memslot(const void *slot1, const void *slot2)
672{
673 struct kvm_memory_slot *s1, *s2;
674
675 s1 = (struct kvm_memory_slot *)slot1;
676 s2 = (struct kvm_memory_slot *)slot2;
677
678 if (s1->npages < s2->npages)
679 return 1;
680 if (s1->npages > s2->npages)
681 return -1;
682
683 return 0;
684}
685
686/*
687 * Sort the memslots base on its size, so the larger slots
688 * will get better fit.
689 */
690static void sort_memslots(struct kvm_memslots *slots)
691{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800692 int i;
693
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800694 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
695 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800696
697 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
698 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800699}
700
Stephen Hemminger79408762013-12-29 12:12:29 -0800701static void update_memslots(struct kvm_memslots *slots,
David Matlackee3d1572014-08-18 15:46:06 -0700702 struct kvm_memory_slot *new)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800703{
704 if (new) {
705 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800706 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800707 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800708
Xiao Guangrong28a37542011-11-24 19:04:35 +0800709 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800710 if (new->npages != npages)
711 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800712 }
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800713}
714
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800715static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
716{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800717 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
718
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200719#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800720 valid_flags |= KVM_MEM_READONLY;
721#endif
722
723 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800724 return -EINVAL;
725
726 return 0;
727}
728
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200729static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
730 struct kvm_memslots *slots, struct kvm_memory_slot *new)
731{
732 struct kvm_memslots *old_memslots = kvm->memslots;
733
David Matlackee3d1572014-08-18 15:46:06 -0700734 /*
735 * Set the low bit in the generation, which disables SPTE caching
736 * until the end of synchronize_srcu_expedited.
737 */
738 WARN_ON(old_memslots->generation & 1);
739 slots->generation = old_memslots->generation + 1;
740
741 update_memslots(slots, new);
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200742 rcu_assign_pointer(kvm->memslots, slots);
743 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900744
David Matlackee3d1572014-08-18 15:46:06 -0700745 /*
746 * Increment the new memslot generation a second time. This prevents
747 * vm exits that race with memslot updates from caching a memslot
748 * generation that will (potentially) be valid forever.
749 */
750 slots->generation++;
751
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900752 kvm_arch_memslots_updated(kvm);
753
754 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200755}
756
Avi Kivity6aa8b732006-12-10 02:21:36 -0800757/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800758 * Allocate some memory and give it an address in the guest physical address
759 * space.
760 *
761 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800762 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500763 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800764 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800765int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900766 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800767{
Avi Kivity8234b222010-12-27 12:08:45 +0200768 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800769 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200770 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900771 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800772 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700773 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900774 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800776 r = check_memory_region_flags(mem);
777 if (r)
778 goto out;
779
Avi Kivity6aa8b732006-12-10 02:21:36 -0800780 r = -EINVAL;
781 /* General sanity checks */
782 if (mem->memory_size & (PAGE_SIZE - 1))
783 goto out;
784 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
785 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900786 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900787 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900788 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200789 !access_ok(VERIFY_WRITE,
790 (void __user *)(unsigned long)mem->userspace_addr,
791 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600792 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800793 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800794 goto out;
795 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
796 goto out;
797
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900798 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800799 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
800 npages = mem->memory_size >> PAGE_SHIFT;
801
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900802 if (npages > KVM_MEM_MAX_NR_PAGES)
803 goto out;
804
Avi Kivity6aa8b732006-12-10 02:21:36 -0800805 if (!npages)
806 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
807
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900808 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800809
Avi Kivitye36d96f2010-06-21 10:56:36 +0300810 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800811 new.base_gfn = base_gfn;
812 new.npages = npages;
813 new.flags = mem->flags;
814
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900815 if (npages) {
816 if (!old.npages)
817 change = KVM_MR_CREATE;
818 else { /* Modify an existing slot. */
819 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900820 (npages != old.npages) ||
821 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900822 goto out;
823
824 if (base_gfn != old.base_gfn)
825 change = KVM_MR_MOVE;
826 else if (new.flags != old.flags)
827 change = KVM_MR_FLAGS_ONLY;
828 else { /* Nothing to change. */
829 r = 0;
830 goto out;
831 }
832 }
833 } else if (old.npages) {
834 change = KVM_MR_DELETE;
835 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900836 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800837
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900838 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900839 /* Check for overlaps */
840 r = -EEXIST;
841 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900842 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
843 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900844 continue;
845 if (!((base_gfn + npages <= slot->base_gfn) ||
846 (base_gfn >= slot->base_gfn + slot->npages)))
847 goto out;
848 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800849 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800850
Avi Kivity6aa8b732006-12-10 02:21:36 -0800851 /* Free page dirty bitmap if unneeded */
852 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000853 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800854
855 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900856 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900857 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900858
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530859 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900860 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800861 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200862
Avi Kivity6aa8b732006-12-10 02:21:36 -0800863 /* Allocate page dirty bitmap if needed */
864 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900865 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800866 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800867 }
868
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900869 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100870 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
871 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200872 if (!slots)
873 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800874 slot = id_to_memslot(slots, mem->slot);
875 slot->flags |= KVM_MEMSLOT_INVALID;
876
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200877 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200878
Alex Williamsone40f1932012-12-10 10:32:57 -0700879 /* slot was deleted or moved, clear iommu mapping */
880 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300881 /* From this point no new shadow pages pointing to a deleted,
882 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200883 *
884 * validation of sp->gfn happens in:
885 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
886 * - kvm_is_visible_gfn (mmu_check_roots)
887 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300888 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700889 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200890 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300891
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900892 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200893 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700894 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200895
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200896 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700897 /*
898 * We can re-use the old_memslots from above, the only difference
899 * from the currently installed memslots is the invalid flag. This
900 * will get overwritten by update_memslots anyway.
901 */
902 if (!slots) {
903 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
904 GFP_KERNEL);
905 if (!slots)
906 goto out_free;
907 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200908
909 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900910 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200911 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900912 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200913 }
914
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200915 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900917 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800918
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530919 kvm_free_physmem_slot(kvm, &old, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200920 kfree(old_memslots);
921
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200922 /*
923 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800924 * un-mapped and re-mapped if their base changes. Since base change
925 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800926 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800927 * slots (size changes, userspace addr changes and read-only flag
928 * changes) is disallowed above, so any other attribute changes getting
929 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200930 */
931 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800932 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800933 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800934 }
935
Avi Kivity6aa8b732006-12-10 02:21:36 -0800936 return 0;
937
938out_slots:
939 kfree(slots);
940out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530941 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800942out:
943 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200944}
945EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800946
947int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900948 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800949{
950 int r;
951
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200952 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900953 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200954 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800955 return r;
956}
Izik Eidus210c7c42007-10-24 23:52:57 +0200957EXPORT_SYMBOL_GPL(kvm_set_memory_region);
958
Stephen Hemminger79408762013-12-29 12:12:29 -0800959static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
960 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200961{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700962 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200963 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900964 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965}
966
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800967int kvm_get_dirty_log(struct kvm *kvm,
968 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969{
970 struct kvm_memory_slot *memslot;
971 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900972 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800973 unsigned long any = 0;
974
Avi Kivity6aa8b732006-12-10 02:21:36 -0800975 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700976 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977 goto out;
978
Xiao Guangrong28a37542011-11-24 19:04:35 +0800979 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800980 r = -ENOENT;
981 if (!memslot->dirty_bitmap)
982 goto out;
983
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900984 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800985
Uri Lublincd1a4a92007-02-22 16:43:09 +0200986 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800987 any = memslot->dirty_bitmap[i];
988
989 r = -EFAULT;
990 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
991 goto out;
992
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800993 if (any)
994 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800995
996 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800997out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800998 return r;
999}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301000EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001001
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001002bool kvm_largepages_enabled(void)
1003{
1004 return largepages_enabled;
1005}
1006
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001007void kvm_disable_largepages(void)
1008{
1009 largepages_enabled = false;
1010}
1011EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1012
Gleb Natapov49c77542010-10-18 15:22:23 +02001013struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1014{
1015 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1016}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001017EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001018
Izik Eiduse0d62c72007-10-24 23:57:46 +02001019int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1020{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001021 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001022
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001023 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001024 memslot->flags & KVM_MEMSLOT_INVALID)
1025 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001026
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001027 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001028}
1029EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1030
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001031unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1032{
1033 struct vm_area_struct *vma;
1034 unsigned long addr, size;
1035
1036 size = PAGE_SIZE;
1037
1038 addr = gfn_to_hva(kvm, gfn);
1039 if (kvm_is_error_hva(addr))
1040 return PAGE_SIZE;
1041
1042 down_read(&current->mm->mmap_sem);
1043 vma = find_vma(current->mm, addr);
1044 if (!vma)
1045 goto out;
1046
1047 size = vma_kernel_pagesize(vma);
1048
1049out:
1050 up_read(&current->mm->mmap_sem);
1051
1052 return size;
1053}
1054
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1056{
1057 return slot->flags & KVM_MEM_READONLY;
1058}
1059
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001060static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1061 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001062{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001063 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001064 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001065
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001066 if (memslot_is_readonly(slot) && write)
1067 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001068
1069 if (nr_pages)
1070 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1071
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001072 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001073}
Xiao Guangrong48987782010-08-22 19:11:43 +08001074
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001075static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1076 gfn_t *nr_pages)
1077{
1078 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1079}
1080
1081unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001082 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001083{
1084 return gfn_to_hva_many(slot, gfn, NULL);
1085}
1086EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1087
Xiao Guangrong48987782010-08-22 19:11:43 +08001088unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1089{
Gleb Natapov49c77542010-10-18 15:22:23 +02001090 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001091}
Sheng Yang0d150292008-04-25 21:44:50 +08001092EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001093
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001094/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001095 * If writable is set to false, the hva returned by this function is only
1096 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001097 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001098unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1099 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001100{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001101 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1102
1103 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001104 *writable = !memslot_is_readonly(slot);
1105
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001106 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001107}
1108
Christoffer Dall64d83122014-08-19 12:15:00 +02001109unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1110{
1111 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1112
1113 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1114}
1115
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001116static int kvm_read_hva(void *data, void __user *hva, int len)
1117{
1118 return __copy_from_user(data, hva, len);
1119}
1120
1121static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1122{
1123 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001124}
1125
Geoff Levand39369f72013-04-05 19:20:30 +00001126static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001127 unsigned long start, int write, struct page **page)
1128{
1129 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1130
1131 if (write)
1132 flags |= FOLL_WRITE;
1133
1134 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1135}
1136
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001137int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1138 unsigned long addr, bool write_fault,
1139 struct page **pagep)
1140{
1141 int npages;
1142 int locked = 1;
1143 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1144 (pagep ? FOLL_GET : 0) |
1145 (write_fault ? FOLL_WRITE : 0);
1146
1147 /*
1148 * If retrying the fault, we get here *not* having allowed the filemap
1149 * to wait on the page lock. We should now allow waiting on the IO with
1150 * the mmap semaphore released.
1151 */
1152 down_read(&mm->mmap_sem);
1153 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1154 &locked);
1155 if (!locked) {
Andres Lagar-Cavillabb0ca6a2014-09-25 15:26:50 -07001156 VM_BUG_ON(npages);
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001157
1158 if (!pagep)
1159 return 0;
1160
1161 /*
1162 * The previous call has now waited on the IO. Now we can
1163 * retry and complete. Pass TRIED to ensure we do not re
1164 * schedule async IO (see e.g. filemap_fault).
1165 */
1166 down_read(&mm->mmap_sem);
1167 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1168 pagep, NULL, NULL);
1169 }
1170 up_read(&mm->mmap_sem);
1171 return npages;
1172}
1173
Huang Yingfafc3db2011-01-30 11:15:49 +08001174static inline int check_user_page_hwpoison(unsigned long addr)
1175{
1176 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1177
1178 rc = __get_user_pages(current, current->mm, addr, 1,
1179 flags, NULL, NULL, NULL);
1180 return rc == -EHWPOISON;
1181}
1182
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001183/*
1184 * The atomic path to get the writable pfn which will be stored in @pfn,
1185 * true indicates success, otherwise false is returned.
1186 */
1187static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1188 bool write_fault, bool *writable, pfn_t *pfn)
1189{
1190 struct page *page[1];
1191 int npages;
1192
1193 if (!(async || atomic))
1194 return false;
1195
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001196 /*
1197 * Fast pin a writable pfn only if it is a write fault request
1198 * or the caller allows to map a writable pfn for a read fault
1199 * request.
1200 */
1201 if (!(write_fault || writable))
1202 return false;
1203
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001204 npages = __get_user_pages_fast(addr, 1, 1, page);
1205 if (npages == 1) {
1206 *pfn = page_to_pfn(page[0]);
1207
1208 if (writable)
1209 *writable = true;
1210 return true;
1211 }
1212
1213 return false;
1214}
1215
1216/*
1217 * The slow path to get the pfn of the specified host virtual address,
1218 * 1 indicates success, -errno is returned if error is detected.
1219 */
1220static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1221 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001222{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001223 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001224 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001225
1226 might_sleep();
1227
1228 if (writable)
1229 *writable = write_fault;
1230
1231 if (async) {
1232 down_read(&current->mm->mmap_sem);
1233 npages = get_user_page_nowait(current, current->mm,
1234 addr, write_fault, page);
1235 up_read(&current->mm->mmap_sem);
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -07001236 } else {
1237 /*
1238 * By now we have tried gup_fast, and possibly async_pf, and we
1239 * are certainly not atomic. Time to retry the gup, allowing
1240 * mmap semaphore to be relinquished in the case of IO.
1241 */
1242 npages = kvm_get_user_page_io(current, current->mm, addr,
1243 write_fault, page);
1244 }
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001245 if (npages != 1)
1246 return npages;
1247
1248 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001249 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001250 struct page *wpage[1];
1251
1252 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1253 if (npages == 1) {
1254 *writable = true;
1255 put_page(page[0]);
1256 page[0] = wpage[0];
1257 }
1258
1259 npages = 1;
1260 }
1261 *pfn = page_to_pfn(page[0]);
1262 return npages;
1263}
1264
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001265static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1266{
1267 if (unlikely(!(vma->vm_flags & VM_READ)))
1268 return false;
1269
1270 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1271 return false;
1272
1273 return true;
1274}
1275
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001276/*
1277 * Pin guest page in memory and return its pfn.
1278 * @addr: host virtual address which maps memory to the guest
1279 * @atomic: whether this function can sleep
1280 * @async: whether this function need to wait IO complete if the
1281 * host page is not in the memory
1282 * @write_fault: whether we should get a writable host page
1283 * @writable: whether it allows to map a writable host page for !@write_fault
1284 *
1285 * The function will map a writable host page for these two cases:
1286 * 1): @write_fault = true
1287 * 2): @write_fault = false && @writable, @writable will tell the caller
1288 * whether the mapping is writable.
1289 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001290static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1291 bool write_fault, bool *writable)
1292{
1293 struct vm_area_struct *vma;
1294 pfn_t pfn = 0;
1295 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001296
Gleb Natapovaf585b92010-10-14 11:22:46 +02001297 /* we can do it either atomically or asynchronously, not both */
1298 BUG_ON(atomic && async);
1299
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001300 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1301 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001302
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001303 if (atomic)
1304 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001305
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001306 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1307 if (npages == 1)
1308 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001309
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001310 down_read(&current->mm->mmap_sem);
1311 if (npages == -EHWPOISON ||
1312 (!async && check_user_page_hwpoison(addr))) {
1313 pfn = KVM_PFN_ERR_HWPOISON;
1314 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001315 }
Izik Eidus539cb662007-11-11 22:05:04 +02001316
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001317 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001318
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001319 if (vma == NULL)
1320 pfn = KVM_PFN_ERR_FAULT;
1321 else if ((vma->vm_flags & VM_PFNMAP)) {
1322 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1323 vma->vm_pgoff;
1324 BUG_ON(!kvm_is_mmio_pfn(pfn));
1325 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001326 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001327 *async = true;
1328 pfn = KVM_PFN_ERR_FAULT;
1329 }
1330exit:
1331 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001332 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001333}
1334
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001335static pfn_t
1336__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1337 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001338{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001339 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1340
1341 if (addr == KVM_HVA_ERR_RO_BAD)
1342 return KVM_PFN_ERR_RO_FAULT;
1343
1344 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001345 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001346
1347 /* Do not map writable pfn in the readonly memslot. */
1348 if (writable && memslot_is_readonly(slot)) {
1349 *writable = false;
1350 writable = NULL;
1351 }
1352
1353 return hva_to_pfn(addr, atomic, async, write_fault,
1354 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001355}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001356
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001357static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1358 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001359{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001360 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001361
Gleb Natapovaf585b92010-10-14 11:22:46 +02001362 if (async)
1363 *async = false;
1364
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001365 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001366
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001367 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1368 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001369}
1370
1371pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1372{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001373 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001374}
1375EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1376
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001377pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1378 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001379{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001380 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001381}
1382EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1383
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001384pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1385{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001386 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001387}
Anthony Liguori35149e22008-04-02 14:46:56 -05001388EXPORT_SYMBOL_GPL(gfn_to_pfn);
1389
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001390pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1391 bool *writable)
1392{
1393 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1394}
1395EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1396
Xiao Guangrongd5661042012-07-17 21:56:16 +08001397pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001398{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001399 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001400}
1401
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001402pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1403{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001404 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001405}
1406EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1407
Xiao Guangrong48987782010-08-22 19:11:43 +08001408int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1409 int nr_pages)
1410{
1411 unsigned long addr;
1412 gfn_t entry;
1413
Gleb Natapov49c77542010-10-18 15:22:23 +02001414 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001415 if (kvm_is_error_hva(addr))
1416 return -1;
1417
1418 if (entry < nr_pages)
1419 return 0;
1420
1421 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1422}
1423EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1424
Xiao Guangronga2766322012-07-26 11:58:59 +08001425static struct page *kvm_pfn_to_page(pfn_t pfn)
1426{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001427 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001428 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001429
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001430 if (kvm_is_mmio_pfn(pfn)) {
1431 WARN_ON(1);
1432 return KVM_ERR_PTR_BAD_PAGE;
1433 }
1434
Xiao Guangronga2766322012-07-26 11:58:59 +08001435 return pfn_to_page(pfn);
1436}
1437
Anthony Liguori35149e22008-04-02 14:46:56 -05001438struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1439{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001440 pfn_t pfn;
1441
1442 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001443
Xiao Guangronga2766322012-07-26 11:58:59 +08001444 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001445}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001446
Avi Kivity954bbbc2007-03-30 14:02:32 +03001447EXPORT_SYMBOL_GPL(gfn_to_page);
1448
Izik Eidusb4231d62007-11-20 11:49:33 +02001449void kvm_release_page_clean(struct page *page)
1450{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001451 WARN_ON(is_error_page(page));
1452
Anthony Liguori35149e22008-04-02 14:46:56 -05001453 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001454}
1455EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1456
Anthony Liguori35149e22008-04-02 14:46:56 -05001457void kvm_release_pfn_clean(pfn_t pfn)
1458{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001459 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001460 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001461}
1462EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1463
Izik Eidusb4231d62007-11-20 11:49:33 +02001464void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001465{
Xiao Guangronga2766322012-07-26 11:58:59 +08001466 WARN_ON(is_error_page(page));
1467
Anthony Liguori35149e22008-04-02 14:46:56 -05001468 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001469}
Izik Eidusb4231d62007-11-20 11:49:33 +02001470EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001471
Stephen Hemminger79408762013-12-29 12:12:29 -08001472static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001473{
1474 kvm_set_pfn_dirty(pfn);
1475 kvm_release_pfn_clean(pfn);
1476}
Anthony Liguori35149e22008-04-02 14:46:56 -05001477
1478void kvm_set_pfn_dirty(pfn_t pfn)
1479{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001480 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001481 struct page *page = pfn_to_page(pfn);
1482 if (!PageReserved(page))
1483 SetPageDirty(page);
1484 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001485}
1486EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1487
1488void kvm_set_pfn_accessed(pfn_t pfn)
1489{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001490 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001491 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001492}
1493EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1494
1495void kvm_get_pfn(pfn_t pfn)
1496{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001497 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001498 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001499}
1500EXPORT_SYMBOL_GPL(kvm_get_pfn);
1501
Izik Eidus195aefd2007-10-01 22:14:18 +02001502static int next_segment(unsigned long len, int offset)
1503{
1504 if (len > PAGE_SIZE - offset)
1505 return PAGE_SIZE - offset;
1506 else
1507 return len;
1508}
1509
1510int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1511 int len)
1512{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001513 int r;
1514 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001515
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001516 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001517 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001518 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001519 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001520 if (r)
1521 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001522 return 0;
1523}
1524EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1525
1526int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1527{
1528 gfn_t gfn = gpa >> PAGE_SHIFT;
1529 int seg;
1530 int offset = offset_in_page(gpa);
1531 int ret;
1532
1533 while ((seg = next_segment(len, offset)) != 0) {
1534 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1535 if (ret < 0)
1536 return ret;
1537 offset = 0;
1538 len -= seg;
1539 data += seg;
1540 ++gfn;
1541 }
1542 return 0;
1543}
1544EXPORT_SYMBOL_GPL(kvm_read_guest);
1545
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001546int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1547 unsigned long len)
1548{
1549 int r;
1550 unsigned long addr;
1551 gfn_t gfn = gpa >> PAGE_SHIFT;
1552 int offset = offset_in_page(gpa);
1553
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001554 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001555 if (kvm_is_error_hva(addr))
1556 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001557 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001558 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001559 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001560 if (r)
1561 return -EFAULT;
1562 return 0;
1563}
1564EXPORT_SYMBOL(kvm_read_guest_atomic);
1565
Izik Eidus195aefd2007-10-01 22:14:18 +02001566int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1567 int offset, int len)
1568{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001569 int r;
1570 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001571
Izik Eiduse0506bc2007-11-11 22:10:22 +02001572 addr = gfn_to_hva(kvm, gfn);
1573 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001574 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001575 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001576 if (r)
1577 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001578 mark_page_dirty(kvm, gfn);
1579 return 0;
1580}
1581EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1582
1583int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1584 unsigned long len)
1585{
1586 gfn_t gfn = gpa >> PAGE_SHIFT;
1587 int seg;
1588 int offset = offset_in_page(gpa);
1589 int ret;
1590
1591 while ((seg = next_segment(len, offset)) != 0) {
1592 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1593 if (ret < 0)
1594 return ret;
1595 offset = 0;
1596 len -= seg;
1597 data += seg;
1598 ++gfn;
1599 }
1600 return 0;
1601}
1602
Gleb Natapov49c77542010-10-18 15:22:23 +02001603int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001604 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001605{
1606 struct kvm_memslots *slots = kvm_memslots(kvm);
1607 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001608 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1609 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1610 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1611 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001612
1613 ghc->gpa = gpa;
1614 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001615 ghc->len = len;
1616 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1617 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1618 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001619 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001620 } else {
1621 /*
1622 * If the requested region crosses two memslots, we still
1623 * verify that the entire region is valid here.
1624 */
1625 while (start_gfn <= end_gfn) {
1626 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1627 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1628 &nr_pages_avail);
1629 if (kvm_is_error_hva(ghc->hva))
1630 return -EFAULT;
1631 start_gfn += nr_pages_avail;
1632 }
1633 /* Use the slow path for cross page reads and writes. */
1634 ghc->memslot = NULL;
1635 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001636 return 0;
1637}
1638EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1639
1640int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1641 void *data, unsigned long len)
1642{
1643 struct kvm_memslots *slots = kvm_memslots(kvm);
1644 int r;
1645
Andrew Honig8f964522013-03-29 09:35:21 -07001646 BUG_ON(len > ghc->len);
1647
Gleb Natapov49c77542010-10-18 15:22:23 +02001648 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001649 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1650
1651 if (unlikely(!ghc->memslot))
1652 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001653
1654 if (kvm_is_error_hva(ghc->hva))
1655 return -EFAULT;
1656
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001657 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001658 if (r)
1659 return -EFAULT;
1660 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1661
1662 return 0;
1663}
1664EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1665
Gleb Natapove03b6442011-07-11 15:28:11 -04001666int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1667 void *data, unsigned long len)
1668{
1669 struct kvm_memslots *slots = kvm_memslots(kvm);
1670 int r;
1671
Andrew Honig8f964522013-03-29 09:35:21 -07001672 BUG_ON(len > ghc->len);
1673
Gleb Natapove03b6442011-07-11 15:28:11 -04001674 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001675 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1676
1677 if (unlikely(!ghc->memslot))
1678 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001679
1680 if (kvm_is_error_hva(ghc->hva))
1681 return -EFAULT;
1682
1683 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1684 if (r)
1685 return -EFAULT;
1686
1687 return 0;
1688}
1689EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1690
Izik Eidus195aefd2007-10-01 22:14:18 +02001691int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1692{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001693 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1694
1695 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001696}
1697EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1698
1699int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1700{
1701 gfn_t gfn = gpa >> PAGE_SHIFT;
1702 int seg;
1703 int offset = offset_in_page(gpa);
1704 int ret;
1705
1706 while ((seg = next_segment(len, offset)) != 0) {
1707 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1708 if (ret < 0)
1709 return ret;
1710 offset = 0;
1711 len -= seg;
1712 ++gfn;
1713 }
1714 return 0;
1715}
1716EXPORT_SYMBOL_GPL(kvm_clear_guest);
1717
Stephen Hemminger79408762013-12-29 12:12:29 -08001718static void mark_page_dirty_in_slot(struct kvm *kvm,
1719 struct kvm_memory_slot *memslot,
1720 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001721{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001722 if (memslot && memslot->dirty_bitmap) {
1723 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001724
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001725 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001726 }
1727}
1728
Gleb Natapov49c77542010-10-18 15:22:23 +02001729void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1730{
1731 struct kvm_memory_slot *memslot;
1732
1733 memslot = gfn_to_memslot(kvm, gfn);
1734 mark_page_dirty_in_slot(kvm, memslot, gfn);
1735}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301736EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001737
Eddie Dongb6958ce2007-07-18 12:15:21 +03001738/*
1739 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1740 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001741void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001742{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001743 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001744
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001745 for (;;) {
1746 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001747
Gleb Natapova1b37102009-07-09 15:33:52 +03001748 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001749 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001750 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001751 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001752 if (kvm_cpu_has_pending_timer(vcpu))
1753 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001754 if (signal_pending(current))
1755 break;
1756
Eddie Dongb6958ce2007-07-18 12:15:21 +03001757 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001758 }
1759
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001760 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001761}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301762EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001763
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001764#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001765/*
1766 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1767 */
1768void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1769{
1770 int me;
1771 int cpu = vcpu->cpu;
1772 wait_queue_head_t *wqp;
1773
1774 wqp = kvm_arch_vcpu_wq(vcpu);
1775 if (waitqueue_active(wqp)) {
1776 wake_up_interruptible(wqp);
1777 ++vcpu->stat.halt_wakeup;
1778 }
1779
1780 me = get_cpu();
1781 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1782 if (kvm_arch_vcpu_should_kick(vcpu))
1783 smp_send_reschedule(cpu);
1784 put_cpu();
1785}
Yang Zhanga20ed542013-04-11 19:25:15 +08001786EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001787#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001788
Dan Carpenterfa933842014-05-23 13:20:42 +03001789int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02001790{
1791 struct pid *pid;
1792 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03001793 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001794
1795 rcu_read_lock();
1796 pid = rcu_dereference(target->pid);
1797 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10001798 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001799 rcu_read_unlock();
1800 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301801 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001802 if (task->flags & PF_VCPU) {
1803 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301804 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001805 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301806 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001807 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301808
1809 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001810}
1811EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1812
Raghavendra K T06e48c52012-07-19 15:17:52 +05301813/*
1814 * Helper that checks whether a VCPU is eligible for directed yield.
1815 * Most eligible candidate to yield is decided by following heuristics:
1816 *
1817 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1818 * (preempted lock holder), indicated by @in_spin_loop.
1819 * Set at the beiginning and cleared at the end of interception/PLE handler.
1820 *
1821 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1822 * chance last time (mostly it has become eligible now since we have probably
1823 * yielded to lockholder in last iteration. This is done by toggling
1824 * @dy_eligible each time a VCPU checked for eligibility.)
1825 *
1826 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1827 * to preempted lock-holder could result in wrong VCPU selection and CPU
1828 * burning. Giving priority for a potential lock-holder increases lock
1829 * progress.
1830 *
1831 * Since algorithm is based on heuristics, accessing another VCPU data without
1832 * locking does not harm. It may result in trying to yield to same VCPU, fail
1833 * and continue with next VCPU and so on.
1834 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001835static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301836{
Scott Wood4a55dd72014-01-09 18:43:16 -06001837#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301838 bool eligible;
1839
1840 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02001841 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301842
1843 if (vcpu->spin_loop.in_spin_loop)
1844 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1845
1846 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001847#else
1848 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301849#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001850}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301851
Rik van Riel217ece62011-02-01 09:53:28 -05001852void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001853{
Rik van Riel217ece62011-02-01 09:53:28 -05001854 struct kvm *kvm = me->kvm;
1855 struct kvm_vcpu *vcpu;
1856 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1857 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301858 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001859 int pass;
1860 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001861
Raghavendra K T4c088492012-07-18 19:07:46 +05301862 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001863 /*
1864 * We boost the priority of a VCPU that is runnable but not
1865 * currently running, because it got preempted by something
1866 * else and called schedule in __vcpu_run. Hopefully that
1867 * VCPU is holding the lock that we need and will release it.
1868 * We approximate round-robin by starting at the last boosted VCPU.
1869 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301870 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001871 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001872 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001873 i = last_boosted_vcpu;
1874 continue;
1875 } else if (pass && i > last_boosted_vcpu)
1876 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301877 if (!ACCESS_ONCE(vcpu->preempted))
1878 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001879 if (vcpu == me)
1880 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001881 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001882 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301883 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1884 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301885
1886 yielded = kvm_vcpu_yield_to(vcpu);
1887 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001888 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001889 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301890 } else if (yielded < 0) {
1891 try--;
1892 if (!try)
1893 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001894 }
Rik van Riel217ece62011-02-01 09:53:28 -05001895 }
1896 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301897 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301898
1899 /* Ensure vcpu is not eligible during next spinloop */
1900 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001901}
1902EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1903
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001904static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001905{
1906 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001907 struct page *page;
1908
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001909 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001910 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001911#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001912 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001913 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001914#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001915#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1916 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1917 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1918#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001919 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001920 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001921 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001922 vmf->page = page;
1923 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001924}
1925
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001926static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001927 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001928};
1929
1930static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1931{
1932 vma->vm_ops = &kvm_vcpu_vm_ops;
1933 return 0;
1934}
1935
Avi Kivitybccf2152007-02-21 18:04:26 +02001936static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1937{
1938 struct kvm_vcpu *vcpu = filp->private_data;
1939
Al Viro66c0b392008-04-19 20:33:56 +01001940 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001941 return 0;
1942}
1943
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001944static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001945 .release = kvm_vcpu_release,
1946 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001947#ifdef CONFIG_COMPAT
1948 .compat_ioctl = kvm_vcpu_compat_ioctl,
1949#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001950 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001951 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001952};
1953
1954/*
1955 * Allocates an inode for the vcpu.
1956 */
1957static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1958{
Yann Droneaud24009b02013-08-24 22:14:07 +02001959 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001960}
1961
Avi Kivityc5ea7662007-02-20 18:41:05 +02001962/*
1963 * Creates some virtual cpus. Good luck creating more than one.
1964 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001965static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001966{
1967 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001968 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001969
Andy Honig338c7db2013-11-18 16:09:22 -08001970 if (id >= KVM_MAX_VCPUS)
1971 return -EINVAL;
1972
Gleb Natapov73880c82009-06-09 15:56:28 +03001973 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001974 if (IS_ERR(vcpu))
1975 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001976
Avi Kivity15ad7142007-07-11 18:17:21 +03001977 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1978
Avi Kivity26e52152007-11-20 15:30:24 +02001979 r = kvm_arch_vcpu_setup(vcpu);
1980 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001981 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001982
Shaohua Li11ec2802007-07-23 14:51:37 +08001983 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001984 if (!kvm_vcpu_compatible(vcpu)) {
1985 r = -EINVAL;
1986 goto unlock_vcpu_destroy;
1987 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001988 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1989 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001990 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001991 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001992
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001993 kvm_for_each_vcpu(r, v, kvm)
1994 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001995 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001996 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001997 }
1998
1999 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002000
2001 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01002002 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002003 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03002004 if (r < 0) {
2005 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02002006 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002007 }
2008
2009 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2010 smp_wmb();
2011 atomic_inc(&kvm->online_vcpus);
2012
Gleb Natapov73880c82009-06-09 15:56:28 +03002013 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02002014 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02002015 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002016
Jan Kiszkad7805922011-05-23 10:33:05 +02002017unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03002018 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02002019vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06002020 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002021 return r;
2022}
2023
Avi Kivity1961d272007-03-05 19:46:05 +02002024static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2025{
2026 if (sigset) {
2027 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2028 vcpu->sigset_active = 1;
2029 vcpu->sigset = *sigset;
2030 } else
2031 vcpu->sigset_active = 0;
2032 return 0;
2033}
2034
Avi Kivitybccf2152007-02-21 18:04:26 +02002035static long kvm_vcpu_ioctl(struct file *filp,
2036 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002037{
Avi Kivitybccf2152007-02-21 18:04:26 +02002038 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002039 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002040 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002041 struct kvm_fpu *fpu = NULL;
2042 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002043
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002044 if (vcpu->kvm->mm != current->mm)
2045 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002046
David Matlack2ea75be2014-09-19 16:03:25 -07002047 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2048 return -EINVAL;
2049
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002050#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002051 /*
2052 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2053 * so vcpu_load() would break it.
2054 */
2055 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2056 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2057#endif
2058
2059
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002060 r = vcpu_load(vcpu);
2061 if (r)
2062 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002063 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002064 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002065 r = -EINVAL;
2066 if (arg)
2067 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002068 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002069 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002070 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002071 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002072 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002073
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002074 r = -ENOMEM;
2075 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2076 if (!kvm_regs)
2077 goto out;
2078 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002079 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002080 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002081 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002082 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2083 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002084 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002085out_free1:
2086 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002087 break;
2088 }
2089 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002090 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002091
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002092 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002093 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2094 if (IS_ERR(kvm_regs)) {
2095 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002096 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002097 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002098 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002099 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002100 break;
2101 }
2102 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002103 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2104 r = -ENOMEM;
2105 if (!kvm_sregs)
2106 goto out;
2107 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002108 if (r)
2109 goto out;
2110 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002111 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002112 goto out;
2113 r = 0;
2114 break;
2115 }
2116 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002117 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2118 if (IS_ERR(kvm_sregs)) {
2119 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002120 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002121 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002122 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002123 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002124 break;
2125 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002126 case KVM_GET_MP_STATE: {
2127 struct kvm_mp_state mp_state;
2128
2129 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2130 if (r)
2131 goto out;
2132 r = -EFAULT;
2133 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2134 goto out;
2135 r = 0;
2136 break;
2137 }
2138 case KVM_SET_MP_STATE: {
2139 struct kvm_mp_state mp_state;
2140
2141 r = -EFAULT;
2142 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2143 goto out;
2144 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002145 break;
2146 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002147 case KVM_TRANSLATE: {
2148 struct kvm_translation tr;
2149
2150 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002151 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002152 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002153 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002154 if (r)
2155 goto out;
2156 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002157 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002158 goto out;
2159 r = 0;
2160 break;
2161 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002162 case KVM_SET_GUEST_DEBUG: {
2163 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002164
2165 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002166 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002167 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002168 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002169 break;
2170 }
Avi Kivity1961d272007-03-05 19:46:05 +02002171 case KVM_SET_SIGNAL_MASK: {
2172 struct kvm_signal_mask __user *sigmask_arg = argp;
2173 struct kvm_signal_mask kvm_sigmask;
2174 sigset_t sigset, *p;
2175
2176 p = NULL;
2177 if (argp) {
2178 r = -EFAULT;
2179 if (copy_from_user(&kvm_sigmask, argp,
2180 sizeof kvm_sigmask))
2181 goto out;
2182 r = -EINVAL;
2183 if (kvm_sigmask.len != sizeof sigset)
2184 goto out;
2185 r = -EFAULT;
2186 if (copy_from_user(&sigset, sigmask_arg->sigset,
2187 sizeof sigset))
2188 goto out;
2189 p = &sigset;
2190 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002191 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002192 break;
2193 }
Avi Kivityb8836732007-04-01 16:34:31 +03002194 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002195 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2196 r = -ENOMEM;
2197 if (!fpu)
2198 goto out;
2199 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002200 if (r)
2201 goto out;
2202 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002203 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002204 goto out;
2205 r = 0;
2206 break;
2207 }
2208 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002209 fpu = memdup_user(argp, sizeof(*fpu));
2210 if (IS_ERR(fpu)) {
2211 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002212 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002213 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002214 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002215 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002216 break;
2217 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002218 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002219 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002220 }
2221out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002222 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002223 kfree(fpu);
2224 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002225 return r;
2226}
2227
Alexander Graf1dda6062011-06-08 02:45:37 +02002228#ifdef CONFIG_COMPAT
2229static long kvm_vcpu_compat_ioctl(struct file *filp,
2230 unsigned int ioctl, unsigned long arg)
2231{
2232 struct kvm_vcpu *vcpu = filp->private_data;
2233 void __user *argp = compat_ptr(arg);
2234 int r;
2235
2236 if (vcpu->kvm->mm != current->mm)
2237 return -EIO;
2238
2239 switch (ioctl) {
2240 case KVM_SET_SIGNAL_MASK: {
2241 struct kvm_signal_mask __user *sigmask_arg = argp;
2242 struct kvm_signal_mask kvm_sigmask;
2243 compat_sigset_t csigset;
2244 sigset_t sigset;
2245
2246 if (argp) {
2247 r = -EFAULT;
2248 if (copy_from_user(&kvm_sigmask, argp,
2249 sizeof kvm_sigmask))
2250 goto out;
2251 r = -EINVAL;
2252 if (kvm_sigmask.len != sizeof csigset)
2253 goto out;
2254 r = -EFAULT;
2255 if (copy_from_user(&csigset, sigmask_arg->sigset,
2256 sizeof csigset))
2257 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002258 sigset_from_compat(&sigset, &csigset);
2259 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2260 } else
2261 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002262 break;
2263 }
2264 default:
2265 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2266 }
2267
2268out:
2269 return r;
2270}
2271#endif
2272
Scott Wood852b6d52013-04-12 14:08:42 +00002273static int kvm_device_ioctl_attr(struct kvm_device *dev,
2274 int (*accessor)(struct kvm_device *dev,
2275 struct kvm_device_attr *attr),
2276 unsigned long arg)
2277{
2278 struct kvm_device_attr attr;
2279
2280 if (!accessor)
2281 return -EPERM;
2282
2283 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2284 return -EFAULT;
2285
2286 return accessor(dev, &attr);
2287}
2288
2289static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2290 unsigned long arg)
2291{
2292 struct kvm_device *dev = filp->private_data;
2293
2294 switch (ioctl) {
2295 case KVM_SET_DEVICE_ATTR:
2296 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2297 case KVM_GET_DEVICE_ATTR:
2298 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2299 case KVM_HAS_DEVICE_ATTR:
2300 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2301 default:
2302 if (dev->ops->ioctl)
2303 return dev->ops->ioctl(dev, ioctl, arg);
2304
2305 return -ENOTTY;
2306 }
2307}
2308
Scott Wood852b6d52013-04-12 14:08:42 +00002309static int kvm_device_release(struct inode *inode, struct file *filp)
2310{
2311 struct kvm_device *dev = filp->private_data;
2312 struct kvm *kvm = dev->kvm;
2313
Scott Wood852b6d52013-04-12 14:08:42 +00002314 kvm_put_kvm(kvm);
2315 return 0;
2316}
2317
2318static const struct file_operations kvm_device_fops = {
2319 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002320#ifdef CONFIG_COMPAT
2321 .compat_ioctl = kvm_device_ioctl,
2322#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002323 .release = kvm_device_release,
2324};
2325
2326struct kvm_device *kvm_device_from_filp(struct file *filp)
2327{
2328 if (filp->f_op != &kvm_device_fops)
2329 return NULL;
2330
2331 return filp->private_data;
2332}
2333
Will Deacond60eacb2014-09-02 10:27:33 +01002334static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2335#ifdef CONFIG_KVM_MPIC
2336 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2337 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2338#endif
2339
2340#ifdef CONFIG_KVM_XICS
2341 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2342#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002343};
2344
2345int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2346{
2347 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2348 return -ENOSPC;
2349
2350 if (kvm_device_ops_table[type] != NULL)
2351 return -EEXIST;
2352
2353 kvm_device_ops_table[type] = ops;
2354 return 0;
2355}
2356
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002357void kvm_unregister_device_ops(u32 type)
2358{
2359 if (kvm_device_ops_table[type] != NULL)
2360 kvm_device_ops_table[type] = NULL;
2361}
2362
Scott Wood852b6d52013-04-12 14:08:42 +00002363static int kvm_ioctl_create_device(struct kvm *kvm,
2364 struct kvm_create_device *cd)
2365{
2366 struct kvm_device_ops *ops = NULL;
2367 struct kvm_device *dev;
2368 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2369 int ret;
2370
Will Deacond60eacb2014-09-02 10:27:33 +01002371 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002372 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002373
2374 ops = kvm_device_ops_table[cd->type];
2375 if (ops == NULL)
2376 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002377
2378 if (test)
2379 return 0;
2380
2381 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2382 if (!dev)
2383 return -ENOMEM;
2384
2385 dev->ops = ops;
2386 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002387
2388 ret = ops->create(dev, cd->type);
2389 if (ret < 0) {
2390 kfree(dev);
2391 return ret;
2392 }
2393
Yann Droneaud24009b02013-08-24 22:14:07 +02002394 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002395 if (ret < 0) {
2396 ops->destroy(dev);
2397 return ret;
2398 }
2399
Scott Wood07f0a7b2013-04-25 14:11:23 +00002400 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002401 kvm_get_kvm(kvm);
2402 cd->fd = ret;
2403 return 0;
2404}
2405
Alexander Graf92b591a2014-07-14 18:33:08 +02002406static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2407{
2408 switch (arg) {
2409 case KVM_CAP_USER_MEMORY:
2410 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2411 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2412#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2413 case KVM_CAP_SET_BOOT_CPU_ID:
2414#endif
2415 case KVM_CAP_INTERNAL_ERROR_DATA:
2416#ifdef CONFIG_HAVE_KVM_MSI
2417 case KVM_CAP_SIGNAL_MSI:
2418#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002419#ifdef CONFIG_HAVE_KVM_IRQFD
Alexander Graf92b591a2014-07-14 18:33:08 +02002420 case KVM_CAP_IRQFD_RESAMPLE:
2421#endif
2422 case KVM_CAP_CHECK_EXTENSION_VM:
2423 return 1;
2424#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2425 case KVM_CAP_IRQ_ROUTING:
2426 return KVM_MAX_IRQ_ROUTES;
2427#endif
2428 default:
2429 break;
2430 }
2431 return kvm_vm_ioctl_check_extension(kvm, arg);
2432}
2433
Avi Kivitybccf2152007-02-21 18:04:26 +02002434static long kvm_vm_ioctl(struct file *filp,
2435 unsigned int ioctl, unsigned long arg)
2436{
2437 struct kvm *kvm = filp->private_data;
2438 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002439 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002440
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002441 if (kvm->mm != current->mm)
2442 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002443 switch (ioctl) {
2444 case KVM_CREATE_VCPU:
2445 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002446 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002447 case KVM_SET_USER_MEMORY_REGION: {
2448 struct kvm_userspace_memory_region kvm_userspace_mem;
2449
2450 r = -EFAULT;
2451 if (copy_from_user(&kvm_userspace_mem, argp,
2452 sizeof kvm_userspace_mem))
2453 goto out;
2454
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002455 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002456 break;
2457 }
2458 case KVM_GET_DIRTY_LOG: {
2459 struct kvm_dirty_log log;
2460
2461 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002462 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002463 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002464 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002465 break;
2466 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002467#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2468 case KVM_REGISTER_COALESCED_MMIO: {
2469 struct kvm_coalesced_mmio_zone zone;
2470 r = -EFAULT;
2471 if (copy_from_user(&zone, argp, sizeof zone))
2472 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002473 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002474 break;
2475 }
2476 case KVM_UNREGISTER_COALESCED_MMIO: {
2477 struct kvm_coalesced_mmio_zone zone;
2478 r = -EFAULT;
2479 if (copy_from_user(&zone, argp, sizeof zone))
2480 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002481 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002482 break;
2483 }
2484#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002485 case KVM_IRQFD: {
2486 struct kvm_irqfd data;
2487
2488 r = -EFAULT;
2489 if (copy_from_user(&data, argp, sizeof data))
2490 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002491 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002492 break;
2493 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002494 case KVM_IOEVENTFD: {
2495 struct kvm_ioeventfd data;
2496
2497 r = -EFAULT;
2498 if (copy_from_user(&data, argp, sizeof data))
2499 goto out;
2500 r = kvm_ioeventfd(kvm, &data);
2501 break;
2502 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002503#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2504 case KVM_SET_BOOT_CPU_ID:
2505 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002506 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002507 if (atomic_read(&kvm->online_vcpus) != 0)
2508 r = -EBUSY;
2509 else
2510 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002511 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002512 break;
2513#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002514#ifdef CONFIG_HAVE_KVM_MSI
2515 case KVM_SIGNAL_MSI: {
2516 struct kvm_msi msi;
2517
2518 r = -EFAULT;
2519 if (copy_from_user(&msi, argp, sizeof msi))
2520 goto out;
2521 r = kvm_send_userspace_msi(kvm, &msi);
2522 break;
2523 }
2524#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002525#ifdef __KVM_HAVE_IRQ_LINE
2526 case KVM_IRQ_LINE_STATUS:
2527 case KVM_IRQ_LINE: {
2528 struct kvm_irq_level irq_event;
2529
2530 r = -EFAULT;
2531 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2532 goto out;
2533
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002534 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2535 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002536 if (r)
2537 goto out;
2538
2539 r = -EFAULT;
2540 if (ioctl == KVM_IRQ_LINE_STATUS) {
2541 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2542 goto out;
2543 }
2544
2545 r = 0;
2546 break;
2547 }
2548#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002549#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2550 case KVM_SET_GSI_ROUTING: {
2551 struct kvm_irq_routing routing;
2552 struct kvm_irq_routing __user *urouting;
2553 struct kvm_irq_routing_entry *entries;
2554
2555 r = -EFAULT;
2556 if (copy_from_user(&routing, argp, sizeof(routing)))
2557 goto out;
2558 r = -EINVAL;
2559 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2560 goto out;
2561 if (routing.flags)
2562 goto out;
2563 r = -ENOMEM;
2564 entries = vmalloc(routing.nr * sizeof(*entries));
2565 if (!entries)
2566 goto out;
2567 r = -EFAULT;
2568 urouting = argp;
2569 if (copy_from_user(entries, urouting->entries,
2570 routing.nr * sizeof(*entries)))
2571 goto out_free_irq_routing;
2572 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2573 routing.flags);
2574 out_free_irq_routing:
2575 vfree(entries);
2576 break;
2577 }
2578#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002579 case KVM_CREATE_DEVICE: {
2580 struct kvm_create_device cd;
2581
2582 r = -EFAULT;
2583 if (copy_from_user(&cd, argp, sizeof(cd)))
2584 goto out;
2585
2586 r = kvm_ioctl_create_device(kvm, &cd);
2587 if (r)
2588 goto out;
2589
2590 r = -EFAULT;
2591 if (copy_to_user(argp, &cd, sizeof(cd)))
2592 goto out;
2593
2594 r = 0;
2595 break;
2596 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002597 case KVM_CHECK_EXTENSION:
2598 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2599 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002600 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002601 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002602 if (r == -ENOTTY)
2603 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002604 }
2605out:
2606 return r;
2607}
2608
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002609#ifdef CONFIG_COMPAT
2610struct compat_kvm_dirty_log {
2611 __u32 slot;
2612 __u32 padding1;
2613 union {
2614 compat_uptr_t dirty_bitmap; /* one bit per page */
2615 __u64 padding2;
2616 };
2617};
2618
2619static long kvm_vm_compat_ioctl(struct file *filp,
2620 unsigned int ioctl, unsigned long arg)
2621{
2622 struct kvm *kvm = filp->private_data;
2623 int r;
2624
2625 if (kvm->mm != current->mm)
2626 return -EIO;
2627 switch (ioctl) {
2628 case KVM_GET_DIRTY_LOG: {
2629 struct compat_kvm_dirty_log compat_log;
2630 struct kvm_dirty_log log;
2631
2632 r = -EFAULT;
2633 if (copy_from_user(&compat_log, (void __user *)arg,
2634 sizeof(compat_log)))
2635 goto out;
2636 log.slot = compat_log.slot;
2637 log.padding1 = compat_log.padding1;
2638 log.padding2 = compat_log.padding2;
2639 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2640
2641 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002642 break;
2643 }
2644 default:
2645 r = kvm_vm_ioctl(filp, ioctl, arg);
2646 }
2647
2648out:
2649 return r;
2650}
2651#endif
2652
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002653static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002654 .release = kvm_vm_release,
2655 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002656#ifdef CONFIG_COMPAT
2657 .compat_ioctl = kvm_vm_compat_ioctl,
2658#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002659 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002660};
2661
Carsten Ottee08b9632012-01-04 10:25:20 +01002662static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002663{
Heiko Carstensaac87632010-10-27 17:22:10 +02002664 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002665 struct kvm *kvm;
2666
Carsten Ottee08b9632012-01-04 10:25:20 +01002667 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002668 if (IS_ERR(kvm))
2669 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002670#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2671 r = kvm_coalesced_mmio_init(kvm);
2672 if (r < 0) {
2673 kvm_put_kvm(kvm);
2674 return r;
2675 }
2676#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002677 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002678 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002679 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002680
Heiko Carstensaac87632010-10-27 17:22:10 +02002681 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002682}
2683
2684static long kvm_dev_ioctl(struct file *filp,
2685 unsigned int ioctl, unsigned long arg)
2686{
Avi Kivity07c45a32007-03-07 13:05:38 +02002687 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002688
2689 switch (ioctl) {
2690 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002691 if (arg)
2692 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002693 r = KVM_API_VERSION;
2694 break;
2695 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002696 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002697 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002698 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02002699 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002700 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002701 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02002702 if (arg)
2703 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002704 r = PAGE_SIZE; /* struct kvm_run */
2705#ifdef CONFIG_X86
2706 r += PAGE_SIZE; /* pio data page */
2707#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002708#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2709 r += PAGE_SIZE; /* coalesced mmio ring page */
2710#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002711 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002712 case KVM_TRACE_ENABLE:
2713 case KVM_TRACE_PAUSE:
2714 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002715 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002716 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002717 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002718 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002719 }
2720out:
2721 return r;
2722}
2723
Avi Kivity6aa8b732006-12-10 02:21:36 -08002724static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002725 .unlocked_ioctl = kvm_dev_ioctl,
2726 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002727 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002728};
2729
2730static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002731 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002732 "kvm",
2733 &kvm_chardev_ops,
2734};
2735
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002736static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002737{
2738 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002739 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002740
Rusty Russell7f59f492008-12-07 21:25:45 +10302741 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002742 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002743
Rusty Russell7f59f492008-12-07 21:25:45 +10302744 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002745
Radim Krčmář13a34e02014-08-28 15:13:03 +02002746 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02002747
2748 if (r) {
2749 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2750 atomic_inc(&hardware_enable_failed);
2751 printk(KERN_INFO "kvm: enabling virtualization on "
2752 "CPU%d failed\n", cpu);
2753 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002754}
2755
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002756static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002757{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002758 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002759 if (kvm_usage_count)
2760 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002761 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002762}
2763
2764static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002765{
2766 int cpu = raw_smp_processor_id();
2767
Rusty Russell7f59f492008-12-07 21:25:45 +10302768 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002769 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302770 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02002771 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03002772}
2773
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002774static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002775{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002776 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002777 if (kvm_usage_count)
2778 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002779 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002780}
2781
Alexander Graf10474ae2009-09-15 11:37:46 +02002782static void hardware_disable_all_nolock(void)
2783{
2784 BUG_ON(!kvm_usage_count);
2785
2786 kvm_usage_count--;
2787 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002788 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002789}
2790
2791static void hardware_disable_all(void)
2792{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002793 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002794 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002795 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002796}
2797
2798static int hardware_enable_all(void)
2799{
2800 int r = 0;
2801
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002802 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002803
2804 kvm_usage_count++;
2805 if (kvm_usage_count == 1) {
2806 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002807 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002808
2809 if (atomic_read(&hardware_enable_failed)) {
2810 hardware_disable_all_nolock();
2811 r = -EBUSY;
2812 }
2813 }
2814
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002815 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002816
2817 return r;
2818}
2819
Avi Kivity774c47f2007-02-12 00:54:47 -08002820static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2821 void *v)
2822{
2823 int cpu = (long)v;
2824
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002825 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002826 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002827 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002828 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2829 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002830 hardware_disable();
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002831 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002832 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002833 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2834 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002835 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002836 break;
2837 }
2838 return NOTIFY_OK;
2839}
2840
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002841static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002842 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002843{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002844 /*
2845 * Some (well, at least mine) BIOSes hang on reboot if
2846 * in vmx root mode.
2847 *
2848 * And Intel TXT required VMX off for all cpu when system shutdown.
2849 */
2850 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2851 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002852 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002853 return NOTIFY_OK;
2854}
2855
2856static struct notifier_block kvm_reboot_notifier = {
2857 .notifier_call = kvm_reboot,
2858 .priority = 0,
2859};
2860
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002861static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002862{
2863 int i;
2864
2865 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002866 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002867
2868 kvm_iodevice_destructor(pos);
2869 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002870 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002871}
2872
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002873static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2874 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002875{
Sasha Levin743eeb02011-07-27 16:00:48 +03002876 if (r1->addr < r2->addr)
2877 return -1;
2878 if (r1->addr + r1->len > r2->addr + r2->len)
2879 return 1;
2880 return 0;
2881}
2882
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002883static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2884{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002885 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002886}
2887
Geoff Levand39369f72013-04-05 19:20:30 +00002888static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002889 gpa_t addr, int len)
2890{
Sasha Levin743eeb02011-07-27 16:00:48 +03002891 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2892 .addr = addr,
2893 .len = len,
2894 .dev = dev,
2895 };
2896
2897 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2898 kvm_io_bus_sort_cmp, NULL);
2899
2900 return 0;
2901}
2902
Geoff Levand39369f72013-04-05 19:20:30 +00002903static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002904 gpa_t addr, int len)
2905{
2906 struct kvm_io_range *range, key;
2907 int off;
2908
2909 key = (struct kvm_io_range) {
2910 .addr = addr,
2911 .len = len,
2912 };
2913
2914 range = bsearch(&key, bus->range, bus->dev_count,
2915 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2916 if (range == NULL)
2917 return -ENOENT;
2918
2919 off = range - bus->range;
2920
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002921 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002922 off--;
2923
2924 return off;
2925}
2926
Cornelia Huck126a5af2013-07-03 16:30:53 +02002927static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2928 struct kvm_io_range *range, const void *val)
2929{
2930 int idx;
2931
2932 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2933 if (idx < 0)
2934 return -EOPNOTSUPP;
2935
2936 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002937 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002938 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2939 range->len, val))
2940 return idx;
2941 idx++;
2942 }
2943
2944 return -EOPNOTSUPP;
2945}
2946
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002947/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002948int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002949 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002950{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002951 struct kvm_io_bus *bus;
2952 struct kvm_io_range range;
2953 int r;
2954
2955 range = (struct kvm_io_range) {
2956 .addr = addr,
2957 .len = len,
2958 };
2959
2960 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2961 r = __kvm_io_bus_write(bus, &range, val);
2962 return r < 0 ? r : 0;
2963}
2964
2965/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2966int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2967 int len, const void *val, long cookie)
2968{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002969 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002970 struct kvm_io_range range;
2971
2972 range = (struct kvm_io_range) {
2973 .addr = addr,
2974 .len = len,
2975 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002976
2977 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002978
2979 /* First try the device referenced by cookie. */
2980 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002981 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002982 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2983 val))
2984 return cookie;
2985
2986 /*
2987 * cookie contained garbage; fall back to search and return the
2988 * correct cookie value.
2989 */
2990 return __kvm_io_bus_write(bus, &range, val);
2991}
2992
2993static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2994 void *val)
2995{
2996 int idx;
2997
2998 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002999 if (idx < 0)
3000 return -EOPNOTSUPP;
3001
3002 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003003 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02003004 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3005 range->len, val))
3006 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03003007 idx++;
3008 }
3009
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003010 return -EOPNOTSUPP;
3011}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03003012EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003013
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003014/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003015int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3016 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003017{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003018 struct kvm_io_bus *bus;
3019 struct kvm_io_range range;
3020 int r;
3021
3022 range = (struct kvm_io_range) {
3023 .addr = addr,
3024 .len = len,
3025 };
3026
3027 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3028 r = __kvm_io_bus_read(bus, &range, val);
3029 return r < 0 ? r : 0;
3030}
3031
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003032
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003033/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003034int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3035 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003036{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003037 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003038
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003039 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08003040 /* exclude ioeventfd which is limited by maximum fd */
3041 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003042 return -ENOSPC;
3043
Amos Konga13007162012-03-09 12:17:32 +08003044 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3045 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003046 if (!new_bus)
3047 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003048 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3049 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003050 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003051 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3052 synchronize_srcu_expedited(&kvm->srcu);
3053 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003054
3055 return 0;
3056}
3057
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003058/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003059int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3060 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003061{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003062 int i, r;
3063 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003064
Sasha Levincdfca7b2011-12-04 19:36:28 +02003065 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003066 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003067 for (i = 0; i < bus->dev_count; i++)
3068 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003069 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003070 break;
3071 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003072
Amos Konga13007162012-03-09 12:17:32 +08003073 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003074 return r;
Amos Konga13007162012-03-09 12:17:32 +08003075
3076 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3077 sizeof(struct kvm_io_range)), GFP_KERNEL);
3078 if (!new_bus)
3079 return -ENOMEM;
3080
3081 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3082 new_bus->dev_count--;
3083 memcpy(new_bus->range + i, bus->range + i + 1,
3084 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003085
3086 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3087 synchronize_srcu_expedited(&kvm->srcu);
3088 kfree(bus);
3089 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003090}
3091
Avi Kivity774c47f2007-02-12 00:54:47 -08003092static struct notifier_block kvm_cpu_notifier = {
3093 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003094};
3095
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003096static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003097{
3098 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003099 struct kvm *kvm;
3100
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003101 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003102 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003103 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003104 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003105 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003106 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003107}
3108
3109DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3110
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003111static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003112{
3113 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003114 struct kvm *kvm;
3115 struct kvm_vcpu *vcpu;
3116 int i;
3117
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003118 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003119 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003120 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003121 kvm_for_each_vcpu(i, vcpu, kvm)
3122 *val += *(u32 *)((void *)vcpu + offset);
3123
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003124 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003125 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003126}
3127
Avi Kivityba1389b2007-11-18 16:24:12 +02003128DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3129
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003130static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003131 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3132 [KVM_STAT_VM] = &vm_stat_fops,
3133};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003134
Hamo4f69b682011-12-15 14:23:16 +08003135static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003136{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003137 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003138 struct kvm_stats_debugfs_item *p;
3139
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003140 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003141 if (kvm_debugfs_dir == NULL)
3142 goto out;
3143
3144 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003145 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003146 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003147 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003148 if (p->dentry == NULL)
3149 goto out_dir;
3150 }
3151
3152 return 0;
3153
3154out_dir:
3155 debugfs_remove_recursive(kvm_debugfs_dir);
3156out:
3157 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003158}
3159
3160static void kvm_exit_debug(void)
3161{
3162 struct kvm_stats_debugfs_item *p;
3163
3164 for (p = debugfs_entries; p->name; ++p)
3165 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003166 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003167}
3168
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003169static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003170{
Alexander Graf10474ae2009-09-15 11:37:46 +02003171 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003172 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003173 return 0;
3174}
3175
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003176static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003177{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003178 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003179 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003180 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003181 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003182}
3183
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003184static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003185 .suspend = kvm_suspend,
3186 .resume = kvm_resume,
3187};
3188
Avi Kivity15ad7142007-07-11 18:17:21 +03003189static inline
3190struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3191{
3192 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3193}
3194
3195static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3196{
3197 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303198 if (vcpu->preempted)
3199 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003200
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003201 kvm_arch_sched_in(vcpu, cpu);
3202
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003203 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003204}
3205
3206static void kvm_sched_out(struct preempt_notifier *pn,
3207 struct task_struct *next)
3208{
3209 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3210
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303211 if (current->state == TASK_RUNNING)
3212 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003213 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003214}
3215
Avi Kivity0ee75be2010-04-28 15:39:01 +03003216int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003217 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003218{
3219 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003220 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003221
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003222 r = kvm_arch_init(opaque);
3223 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003224 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003225
Asias He7dac16c2013-05-08 10:57:29 +08003226 /*
3227 * kvm_arch_init makes sure there's at most one caller
3228 * for architectures that support multiple implementations,
3229 * like intel and amd on x86.
3230 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3231 * conflicts in case kvm is already setup for another implementation.
3232 */
3233 r = kvm_irqfd_init();
3234 if (r)
3235 goto out_irqfd;
3236
Avi Kivity8437a612009-06-06 14:52:35 -07003237 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303238 r = -ENOMEM;
3239 goto out_free_0;
3240 }
3241
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003242 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003243 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303244 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003245
Yang, Sheng002c7f72007-07-31 14:23:01 +03003246 for_each_online_cpu(cpu) {
3247 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003248 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003249 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003250 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003251 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003252 }
3253
Avi Kivity774c47f2007-02-12 00:54:47 -08003254 r = register_cpu_notifier(&kvm_cpu_notifier);
3255 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003256 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003257 register_reboot_notifier(&kvm_reboot_notifier);
3258
Rusty Russellc16f8622007-07-30 21:12:19 +10003259 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003260 if (!vcpu_align)
3261 vcpu_align = __alignof__(struct kvm_vcpu);
3262 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003263 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003264 if (!kvm_vcpu_cache) {
3265 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003266 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003267 }
3268
Gleb Natapovaf585b92010-10-14 11:22:46 +02003269 r = kvm_async_pf_init();
3270 if (r)
3271 goto out_free;
3272
Avi Kivity6aa8b732006-12-10 02:21:36 -08003273 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003274 kvm_vm_fops.owner = module;
3275 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003276
3277 r = misc_register(&kvm_dev);
3278 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003279 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003280 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003281 }
3282
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003283 register_syscore_ops(&kvm_syscore_ops);
3284
Avi Kivity15ad7142007-07-11 18:17:21 +03003285 kvm_preempt_ops.sched_in = kvm_sched_in;
3286 kvm_preempt_ops.sched_out = kvm_sched_out;
3287
Hamo4f69b682011-12-15 14:23:16 +08003288 r = kvm_init_debug();
3289 if (r) {
3290 printk(KERN_ERR "kvm: create debugfs files failed\n");
3291 goto out_undebugfs;
3292 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003293
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003294 r = kvm_vfio_ops_init();
3295 WARN_ON(r);
3296
Avi Kivityc7addb92007-09-16 18:58:32 +02003297 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003298
Hamo4f69b682011-12-15 14:23:16 +08003299out_undebugfs:
3300 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003301 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003302out_unreg:
3303 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003304out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003305 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003306out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003307 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003308 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003309out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003310out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003311 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303312out_free_0a:
3313 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003314out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003315 kvm_irqfd_exit();
3316out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003317 kvm_arch_exit();
3318out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003319 return r;
3320}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003321EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003322
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003323void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003324{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003325 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003326 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003327 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003328 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003329 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003330 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003331 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003332 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003333 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003334 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003335 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303336 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08003337 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003338}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003339EXPORT_SYMBOL_GPL(kvm_exit);