blob: 33712fb26eb11caa0f14bbf3b21a9e459c18252a [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Paolo Bonzini2f303b72013-09-25 13:53:07 +020073DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020074static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080075LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080076
Rusty Russell7f59f492008-12-07 21:25:45 +103077static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020078static int kvm_usage_count = 0;
79static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030080
Rusty Russellc16f8622007-07-30 21:12:19 +100081struct kmem_cache *kvm_vcpu_cache;
82EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030083
Avi Kivity15ad7142007-07-11 18:17:21 +030084static __read_mostly struct preempt_ops kvm_preempt_ops;
85
Hollis Blanchard76f7c872008-04-15 16:05:42 -050086struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080087
Avi Kivitybccf2152007-02-21 18:04:26 +020088static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
89 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020090#ifdef CONFIG_COMPAT
91static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
92 unsigned long arg);
93#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020094static int hardware_enable_all(void);
95static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020096
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020097static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -080098static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
100
101static void kvm_release_pfn_dirty(pfn_t pfn);
102static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200104
Andi Kleen52480132014-02-08 08:51:57 +0100105__visible bool kvm_rebooting;
Avi Kivityb7c41452010-12-02 17:52:50 +0200106EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300107
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300108static bool largepages_enabled = true;
109
Xiao Guangronga2766322012-07-26 11:58:59 +0800110bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300111{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200112 if (pfn_valid(pfn))
113 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300114
115 return true;
116}
117
Avi Kivity6aa8b732006-12-10 02:21:36 -0800118/*
119 * Switches to specified vcpu, until a matching vcpu_put()
120 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300121int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800122{
Avi Kivity15ad7142007-07-11 18:17:21 +0300123 int cpu;
124
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300125 if (mutex_lock_killable(&vcpu->mutex))
126 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
132 synchronize_rcu();
133 put_pid(oldpid);
134 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300135 cpu = get_cpu();
136 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200137 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300138 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300139 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200140}
141
Carsten Otte313a3dc2007-10-11 19:16:52 +0200142void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800143{
Avi Kivity15ad7142007-07-11 18:17:21 +0300144 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200145 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300146 preempt_notifier_unregister(&vcpu->preempt_notifier);
147 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800148 mutex_unlock(&vcpu->mutex);
149}
150
Avi Kivityd9e368d2007-06-07 19:18:30 +0300151static void ack_flush(void *_completed)
152{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300153}
154
Rusty Russell49846892008-12-08 20:26:24 +1030155static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300156{
Avi Kivity597a5f52008-07-20 14:24:22 +0300157 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030158 cpumask_var_t cpus;
159 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300160 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300161
Li Zefan79f55992009-06-15 14:58:26 +0800162 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030163
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800164 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300165 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800166 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300167 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800168
169 /* Set ->requests bit before we read ->mode */
170 smp_mb();
171
172 if (cpus != NULL && cpu != -1 && cpu != me &&
173 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030174 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300175 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176 if (unlikely(cpus == NULL))
177 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
178 else if (!cpumask_empty(cpus))
179 smp_call_function_many(cpus, ack_flush, NULL, 1);
180 else
181 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800182 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030183 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030184 return called;
185}
186
187void kvm_flush_remote_tlbs(struct kvm *kvm)
188{
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800189 long dirty_count = kvm->tlbs_dirty;
190
191 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030192 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
193 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800194 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300195}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530196EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300197
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500198void kvm_reload_remote_mmus(struct kvm *kvm)
199{
Rusty Russell49846892008-12-08 20:26:24 +1030200 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500201}
202
Marcelo Tosattid8281992012-11-27 23:29:01 -0200203void kvm_make_mclock_inprogress_request(struct kvm *kvm)
204{
205 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
206}
207
Yang Zhang3d81bc72013-04-11 19:25:13 +0800208void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800209{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800210 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800211}
212
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000213int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214{
215 struct page *page;
216 int r;
217
218 mutex_init(&vcpu->mutex);
219 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000220 vcpu->kvm = kvm;
221 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500222 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300223 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200224 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000225
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227 if (!page) {
228 r = -ENOMEM;
229 goto fail;
230 }
231 vcpu->run = page_address(page);
232
Raghavendra K T4c088492012-07-18 19:07:46 +0530233 kvm_vcpu_set_in_spin_loop(vcpu, false);
234 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530235 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530236
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800237 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000238 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800239 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000240 return 0;
241
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000242fail_free_run:
243 free_page((unsigned long)vcpu->run);
244fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000245 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000246}
247EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248
249void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
250{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500251 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800252 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000253 free_page((unsigned long)vcpu->run);
254}
255EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200257#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259{
260 return container_of(mn, struct kvm, mmu_notifier);
261}
262
263static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 struct mm_struct *mm,
265 unsigned long address)
266{
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200268 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200269
270 /*
271 * When ->invalidate_page runs, the linux pte has been zapped
272 * already but the page is still allocated until
273 * ->invalidate_page returns. So if we increase the sequence
274 * here the kvm page fault will notice if the spte can't be
275 * established because the page is going to be freed. If
276 * instead the kvm page fault establishes the spte before
277 * ->invalidate_page runs, kvm_unmap_hva will release it
278 * before returning.
279 *
280 * The sequence increase only need to be seen at spin_unlock
281 * time, and not at spin_lock time.
282 *
283 * Increasing the sequence after the spin_unlock would be
284 * unsafe because the kvm page fault could then establish the
285 * pte after kvm_unmap_hva returned, without noticing the page
286 * is going to be freed.
287 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200288 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200289 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900290
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200291 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800292 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200293 /* we've to flush the tlb before the pages can be freed */
294 if (need_tlb_flush)
295 kvm_flush_remote_tlbs(kvm);
296
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900297 spin_unlock(&kvm->mmu_lock);
298 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200299}
300
Izik Eidus3da0dd42009-09-23 21:47:18 +0300301static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
304 pte_t pte)
305{
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200307 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300308
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200309 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200314 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300315}
316
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200317static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
319 unsigned long start,
320 unsigned long end)
321{
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200323 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200324
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200325 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200326 spin_lock(&kvm->mmu_lock);
327 /*
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
331 */
332 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800334 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200335 /* we've to flush the tlb before the pages can be freed */
336 if (need_tlb_flush)
337 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900338
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200341}
342
343static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
345 unsigned long start,
346 unsigned long end)
347{
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349
350 spin_lock(&kvm->mmu_lock);
351 /*
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
354 * been freed.
355 */
356 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000357 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200358 /*
359 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200362 */
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
365
366 BUG_ON(kvm->mmu_notifier_count < 0);
367}
368
369static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
371 unsigned long address)
372{
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200374 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200375
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200376 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200377 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200378
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900379 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200380 if (young)
381 kvm_flush_remote_tlbs(kvm);
382
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900383 spin_unlock(&kvm->mmu_lock);
384 srcu_read_unlock(&kvm->srcu, idx);
385
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200386 return young;
387}
388
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800389static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
390 struct mm_struct *mm,
391 unsigned long address)
392{
393 struct kvm *kvm = mmu_notifier_to_kvm(mn);
394 int young, idx;
395
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
398 young = kvm_test_age_hva(kvm, address);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
401
402 return young;
403}
404
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100405static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
406 struct mm_struct *mm)
407{
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800409 int idx;
410
411 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300412 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800413 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100414}
415
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200416static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
417 .invalidate_page = kvm_mmu_notifier_invalidate_page,
418 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
419 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
420 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800421 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300422 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100423 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200424};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200425
426static int kvm_init_mmu_notifier(struct kvm *kvm)
427{
428 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
429 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
430}
431
432#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
433
434static int kvm_init_mmu_notifier(struct kvm *kvm)
435{
436 return 0;
437}
438
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200439#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
440
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800441static void kvm_init_memslots_id(struct kvm *kvm)
442{
443 int i;
444 struct kvm_memslots *slots = kvm->memslots;
445
446 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800447 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800448}
449
Carsten Ottee08b9632012-01-04 10:25:20 +0100450static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800451{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100452 int r, i;
453 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800454
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100455 if (!kvm)
456 return ERR_PTR(-ENOMEM);
457
Carsten Ottee08b9632012-01-04 10:25:20 +0100458 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100459 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100460 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200461
462 r = hardware_enable_all();
463 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100464 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200465
Avi Kivity75858a82009-01-04 17:10:50 +0200466#ifdef CONFIG_HAVE_KVM_IRQCHIP
467 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200468#endif
469#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300470 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200471#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800472
Alex Williamson1e702d92012-12-10 10:33:32 -0700473 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
474
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200475 r = -ENOMEM;
476 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
477 if (!kvm->memslots)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100478 goto out_err_no_srcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800479 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200480 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100481 goto out_err_no_srcu;
482 if (init_srcu_struct(&kvm->irq_srcu))
483 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
486 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100487 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200488 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200489 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700490
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900491 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400494 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800495 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300496 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200497 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300498 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000499 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700500
501 r = kvm_init_mmu_notifier(kvm);
502 if (r)
503 goto out_err;
504
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200505 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000506 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200507 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100508
Avi Kivityf17abe92007-02-21 19:28:04 +0200509 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200510
511out_err:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100512 cleanup_srcu_struct(&kvm->irq_srcu);
513out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100514 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100515out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200516 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100517out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200518 for (i = 0; i < KVM_NR_BUSES; i++)
519 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200520 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100521 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200522 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200523}
524
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900525/*
526 * Avoid using vmalloc for a small buffer.
527 * Should not be used when the size is statically known.
528 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900529void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900530{
531 if (size > PAGE_SIZE)
532 return vzalloc(size);
533 else
534 return kzalloc(size, GFP_KERNEL);
535}
536
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900537void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900538{
539 if (is_vmalloc_addr(addr))
540 vfree(addr);
541 else
542 kfree(addr);
543}
544
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900545static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
546{
547 if (!memslot->dirty_bitmap)
548 return;
549
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900550 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900551 memslot->dirty_bitmap = NULL;
552}
553
Avi Kivity6aa8b732006-12-10 02:21:36 -0800554/*
555 * Free any memory in @free but not in @dont.
556 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530557static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800558 struct kvm_memory_slot *dont)
559{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900561 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800562
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530563 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300564
Avi Kivity6aa8b732006-12-10 02:21:36 -0800565 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566}
567
Stephen Hemminger79408762013-12-29 12:12:29 -0800568static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200570 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800571 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800572
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800573 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530574 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200575
576 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800577}
578
Scott Wood07f0a7b2013-04-25 14:11:23 +0000579static void kvm_destroy_devices(struct kvm *kvm)
580{
581 struct list_head *node, *tmp;
582
583 list_for_each_safe(node, tmp, &kvm->devices) {
584 struct kvm_device *dev =
585 list_entry(node, struct kvm_device, vm_node);
586
587 list_del(node);
588 dev->ops->destroy(dev);
589 }
590}
591
Avi Kivityf17abe92007-02-21 19:28:04 +0200592static void kvm_destroy_vm(struct kvm *kvm)
593{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200594 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200595 struct mm_struct *mm = kvm->mm;
596
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800597 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200598 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800599 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200600 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200601 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200602 for (i = 0; i < KVM_NR_BUSES; i++)
603 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200604 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200605#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
606 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200607#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300608 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200609#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800610 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000611 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100612 kvm_free_physmem(kvm);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200613 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100614 cleanup_srcu_struct(&kvm->srcu);
615 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200616 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200617 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200618}
619
Izik Eidusd39f13b2008-03-30 16:01:25 +0300620void kvm_get_kvm(struct kvm *kvm)
621{
622 atomic_inc(&kvm->users_count);
623}
624EXPORT_SYMBOL_GPL(kvm_get_kvm);
625
626void kvm_put_kvm(struct kvm *kvm)
627{
628 if (atomic_dec_and_test(&kvm->users_count))
629 kvm_destroy_vm(kvm);
630}
631EXPORT_SYMBOL_GPL(kvm_put_kvm);
632
633
Avi Kivityf17abe92007-02-21 19:28:04 +0200634static int kvm_vm_release(struct inode *inode, struct file *filp)
635{
636 struct kvm *kvm = filp->private_data;
637
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400638 kvm_irqfd_release(kvm);
639
Izik Eidusd39f13b2008-03-30 16:01:25 +0300640 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800641 return 0;
642}
643
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900644/*
645 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900646 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900647 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900648static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
649{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900650 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900651
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900652 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900653 if (!memslot->dirty_bitmap)
654 return -ENOMEM;
655
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900656 return 0;
657}
658
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800659static int cmp_memslot(const void *slot1, const void *slot2)
660{
661 struct kvm_memory_slot *s1, *s2;
662
663 s1 = (struct kvm_memory_slot *)slot1;
664 s2 = (struct kvm_memory_slot *)slot2;
665
666 if (s1->npages < s2->npages)
667 return 1;
668 if (s1->npages > s2->npages)
669 return -1;
670
671 return 0;
672}
673
674/*
675 * Sort the memslots base on its size, so the larger slots
676 * will get better fit.
677 */
678static void sort_memslots(struct kvm_memslots *slots)
679{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800680 int i;
681
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800682 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
683 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800684
685 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
686 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800687}
688
Stephen Hemminger79408762013-12-29 12:12:29 -0800689static void update_memslots(struct kvm_memslots *slots,
690 struct kvm_memory_slot *new,
691 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800692{
693 if (new) {
694 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800695 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800696 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800697
Xiao Guangrong28a37542011-11-24 19:04:35 +0800698 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800699 if (new->npages != npages)
700 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800701 }
702
Alex Williamson116c14c2012-12-21 08:20:16 -0700703 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800704}
705
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800706static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
707{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800708 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
709
710#ifdef KVM_CAP_READONLY_MEM
711 valid_flags |= KVM_MEM_READONLY;
712#endif
713
714 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800715 return -EINVAL;
716
717 return 0;
718}
719
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200720static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
721 struct kvm_memslots *slots, struct kvm_memory_slot *new)
722{
723 struct kvm_memslots *old_memslots = kvm->memslots;
724
725 update_memslots(slots, new, kvm->memslots->generation);
726 rcu_assign_pointer(kvm->memslots, slots);
727 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900728
729 kvm_arch_memslots_updated(kvm);
730
731 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200732}
733
Avi Kivity6aa8b732006-12-10 02:21:36 -0800734/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800735 * Allocate some memory and give it an address in the guest physical address
736 * space.
737 *
738 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800739 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500740 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800741 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800742int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900743 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800744{
Avi Kivity8234b222010-12-27 12:08:45 +0200745 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800746 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200747 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900748 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800749 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700750 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900751 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800752
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800753 r = check_memory_region_flags(mem);
754 if (r)
755 goto out;
756
Avi Kivity6aa8b732006-12-10 02:21:36 -0800757 r = -EINVAL;
758 /* General sanity checks */
759 if (mem->memory_size & (PAGE_SIZE - 1))
760 goto out;
761 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
762 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900763 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900764 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900765 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200766 !access_ok(VERIFY_WRITE,
767 (void __user *)(unsigned long)mem->userspace_addr,
768 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600769 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800770 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800771 goto out;
772 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
773 goto out;
774
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900775 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800776 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
777 npages = mem->memory_size >> PAGE_SHIFT;
778
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900779 r = -EINVAL;
780 if (npages > KVM_MEM_MAX_NR_PAGES)
781 goto out;
782
Avi Kivity6aa8b732006-12-10 02:21:36 -0800783 if (!npages)
784 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
785
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900786 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800787
Avi Kivitye36d96f2010-06-21 10:56:36 +0300788 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800789 new.base_gfn = base_gfn;
790 new.npages = npages;
791 new.flags = mem->flags;
792
Avi Kivity6aa8b732006-12-10 02:21:36 -0800793 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900794 if (npages) {
795 if (!old.npages)
796 change = KVM_MR_CREATE;
797 else { /* Modify an existing slot. */
798 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900799 (npages != old.npages) ||
800 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900801 goto out;
802
803 if (base_gfn != old.base_gfn)
804 change = KVM_MR_MOVE;
805 else if (new.flags != old.flags)
806 change = KVM_MR_FLAGS_ONLY;
807 else { /* Nothing to change. */
808 r = 0;
809 goto out;
810 }
811 }
812 } else if (old.npages) {
813 change = KVM_MR_DELETE;
814 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900815 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800816
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900817 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900818 /* Check for overlaps */
819 r = -EEXIST;
820 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900821 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
822 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900823 continue;
824 if (!((base_gfn + npages <= slot->base_gfn) ||
825 (base_gfn >= slot->base_gfn + slot->npages)))
826 goto out;
827 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800828 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800829
Avi Kivity6aa8b732006-12-10 02:21:36 -0800830 /* Free page dirty bitmap if unneeded */
831 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000832 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800833
834 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900835 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900836 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900837
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530838 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900839 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800840 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200841
Avi Kivity6aa8b732006-12-10 02:21:36 -0800842 /* Allocate page dirty bitmap if needed */
843 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900844 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800845 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800846 }
847
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900848 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200849 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100850 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
851 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200852 if (!slots)
853 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800854 slot = id_to_memslot(slots, mem->slot);
855 slot->flags |= KVM_MEMSLOT_INVALID;
856
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200857 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200858
Alex Williamsone40f1932012-12-10 10:32:57 -0700859 /* slot was deleted or moved, clear iommu mapping */
860 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300861 /* From this point no new shadow pages pointing to a deleted,
862 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200863 *
864 * validation of sp->gfn happens in:
865 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
866 * - kvm_is_visible_gfn (mmu_check_roots)
867 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300868 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700869 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200870 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300871
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900872 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200873 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700874 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200875
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200876 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700877 /*
878 * We can re-use the old_memslots from above, the only difference
879 * from the currently installed memslots is the invalid flag. This
880 * will get overwritten by update_memslots anyway.
881 */
882 if (!slots) {
883 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
884 GFP_KERNEL);
885 if (!slots)
886 goto out_free;
887 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200888
889 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900890 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200891 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900892 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200893 }
894
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200895 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800896
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900897 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800898
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530899 kvm_free_physmem_slot(kvm, &old, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200900 kfree(old_memslots);
901
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200902 /*
903 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800904 * un-mapped and re-mapped if their base changes. Since base change
905 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800906 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800907 * slots (size changes, userspace addr changes and read-only flag
908 * changes) is disallowed above, so any other attribute changes getting
909 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200910 */
911 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800912 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800913 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800914 }
915
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916 return 0;
917
918out_slots:
919 kfree(slots);
920out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530921 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800922out:
923 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200924}
925EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800926
927int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900928 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800929{
930 int r;
931
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200932 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900933 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200934 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800935 return r;
936}
Izik Eidus210c7c42007-10-24 23:52:57 +0200937EXPORT_SYMBOL_GPL(kvm_set_memory_region);
938
Stephen Hemminger79408762013-12-29 12:12:29 -0800939static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
940 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200941{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700942 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200943 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900944 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945}
946
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800947int kvm_get_dirty_log(struct kvm *kvm,
948 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800949{
950 struct kvm_memory_slot *memslot;
951 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900952 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800953 unsigned long any = 0;
954
Avi Kivity6aa8b732006-12-10 02:21:36 -0800955 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700956 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957 goto out;
958
Xiao Guangrong28a37542011-11-24 19:04:35 +0800959 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800960 r = -ENOENT;
961 if (!memslot->dirty_bitmap)
962 goto out;
963
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900964 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965
Uri Lublincd1a4a92007-02-22 16:43:09 +0200966 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967 any = memslot->dirty_bitmap[i];
968
969 r = -EFAULT;
970 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
971 goto out;
972
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800973 if (any)
974 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800975
976 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800978 return r;
979}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530980EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800981
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900982bool kvm_largepages_enabled(void)
983{
984 return largepages_enabled;
985}
986
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300987void kvm_disable_largepages(void)
988{
989 largepages_enabled = false;
990}
991EXPORT_SYMBOL_GPL(kvm_disable_largepages);
992
Gleb Natapov49c77542010-10-18 15:22:23 +0200993struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
994{
995 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
996}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300997EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800998
Izik Eiduse0d62c72007-10-24 23:57:46 +0200999int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1000{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001001 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001002
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001003 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001004 memslot->flags & KVM_MEMSLOT_INVALID)
1005 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001006
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001007 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001008}
1009EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1010
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001011unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1012{
1013 struct vm_area_struct *vma;
1014 unsigned long addr, size;
1015
1016 size = PAGE_SIZE;
1017
1018 addr = gfn_to_hva(kvm, gfn);
1019 if (kvm_is_error_hva(addr))
1020 return PAGE_SIZE;
1021
1022 down_read(&current->mm->mmap_sem);
1023 vma = find_vma(current->mm, addr);
1024 if (!vma)
1025 goto out;
1026
1027 size = vma_kernel_pagesize(vma);
1028
1029out:
1030 up_read(&current->mm->mmap_sem);
1031
1032 return size;
1033}
1034
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001035static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1036{
1037 return slot->flags & KVM_MEM_READONLY;
1038}
1039
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001040static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1041 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001042{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001043 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001044 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001045
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001046 if (memslot_is_readonly(slot) && write)
1047 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001048
1049 if (nr_pages)
1050 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1051
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001052 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001053}
Xiao Guangrong48987782010-08-22 19:11:43 +08001054
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056 gfn_t *nr_pages)
1057{
1058 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1059}
1060
1061unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001062 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001063{
1064 return gfn_to_hva_many(slot, gfn, NULL);
1065}
1066EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1067
Xiao Guangrong48987782010-08-22 19:11:43 +08001068unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1069{
Gleb Natapov49c77542010-10-18 15:22:23 +02001070 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001071}
Sheng Yang0d150292008-04-25 21:44:50 +08001072EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001073
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001074/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001075 * If writable is set to false, the hva returned by this function is only
1076 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001077 */
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001078unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001079{
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001080 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001081 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1082
1083 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001084 *writable = !memslot_is_readonly(slot);
1085
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001086 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001087}
1088
1089static int kvm_read_hva(void *data, void __user *hva, int len)
1090{
1091 return __copy_from_user(data, hva, len);
1092}
1093
1094static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1095{
1096 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001097}
1098
Geoff Levand39369f72013-04-05 19:20:30 +00001099static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001100 unsigned long start, int write, struct page **page)
1101{
1102 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1103
1104 if (write)
1105 flags |= FOLL_WRITE;
1106
1107 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1108}
1109
Huang Yingfafc3db2011-01-30 11:15:49 +08001110static inline int check_user_page_hwpoison(unsigned long addr)
1111{
1112 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1113
1114 rc = __get_user_pages(current, current->mm, addr, 1,
1115 flags, NULL, NULL, NULL);
1116 return rc == -EHWPOISON;
1117}
1118
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001119/*
1120 * The atomic path to get the writable pfn which will be stored in @pfn,
1121 * true indicates success, otherwise false is returned.
1122 */
1123static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1124 bool write_fault, bool *writable, pfn_t *pfn)
1125{
1126 struct page *page[1];
1127 int npages;
1128
1129 if (!(async || atomic))
1130 return false;
1131
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001132 /*
1133 * Fast pin a writable pfn only if it is a write fault request
1134 * or the caller allows to map a writable pfn for a read fault
1135 * request.
1136 */
1137 if (!(write_fault || writable))
1138 return false;
1139
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001140 npages = __get_user_pages_fast(addr, 1, 1, page);
1141 if (npages == 1) {
1142 *pfn = page_to_pfn(page[0]);
1143
1144 if (writable)
1145 *writable = true;
1146 return true;
1147 }
1148
1149 return false;
1150}
1151
1152/*
1153 * The slow path to get the pfn of the specified host virtual address,
1154 * 1 indicates success, -errno is returned if error is detected.
1155 */
1156static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1157 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001158{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001159 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001160 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001161
1162 might_sleep();
1163
1164 if (writable)
1165 *writable = write_fault;
1166
1167 if (async) {
1168 down_read(&current->mm->mmap_sem);
1169 npages = get_user_page_nowait(current, current->mm,
1170 addr, write_fault, page);
1171 up_read(&current->mm->mmap_sem);
1172 } else
1173 npages = get_user_pages_fast(addr, 1, write_fault,
1174 page);
1175 if (npages != 1)
1176 return npages;
1177
1178 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001179 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001180 struct page *wpage[1];
1181
1182 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1183 if (npages == 1) {
1184 *writable = true;
1185 put_page(page[0]);
1186 page[0] = wpage[0];
1187 }
1188
1189 npages = 1;
1190 }
1191 *pfn = page_to_pfn(page[0]);
1192 return npages;
1193}
1194
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001195static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1196{
1197 if (unlikely(!(vma->vm_flags & VM_READ)))
1198 return false;
1199
1200 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1201 return false;
1202
1203 return true;
1204}
1205
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001206/*
1207 * Pin guest page in memory and return its pfn.
1208 * @addr: host virtual address which maps memory to the guest
1209 * @atomic: whether this function can sleep
1210 * @async: whether this function need to wait IO complete if the
1211 * host page is not in the memory
1212 * @write_fault: whether we should get a writable host page
1213 * @writable: whether it allows to map a writable host page for !@write_fault
1214 *
1215 * The function will map a writable host page for these two cases:
1216 * 1): @write_fault = true
1217 * 2): @write_fault = false && @writable, @writable will tell the caller
1218 * whether the mapping is writable.
1219 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001220static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1221 bool write_fault, bool *writable)
1222{
1223 struct vm_area_struct *vma;
1224 pfn_t pfn = 0;
1225 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001226
Gleb Natapovaf585b92010-10-14 11:22:46 +02001227 /* we can do it either atomically or asynchronously, not both */
1228 BUG_ON(atomic && async);
1229
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001230 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1231 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001232
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001233 if (atomic)
1234 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001235
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001236 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1237 if (npages == 1)
1238 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001239
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001240 down_read(&current->mm->mmap_sem);
1241 if (npages == -EHWPOISON ||
1242 (!async && check_user_page_hwpoison(addr))) {
1243 pfn = KVM_PFN_ERR_HWPOISON;
1244 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001245 }
Izik Eidus539cb662007-11-11 22:05:04 +02001246
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001247 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001248
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001249 if (vma == NULL)
1250 pfn = KVM_PFN_ERR_FAULT;
1251 else if ((vma->vm_flags & VM_PFNMAP)) {
1252 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1253 vma->vm_pgoff;
1254 BUG_ON(!kvm_is_mmio_pfn(pfn));
1255 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001256 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001257 *async = true;
1258 pfn = KVM_PFN_ERR_FAULT;
1259 }
1260exit:
1261 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001262 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001263}
1264
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001265static pfn_t
1266__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1267 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001268{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001269 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1270
1271 if (addr == KVM_HVA_ERR_RO_BAD)
1272 return KVM_PFN_ERR_RO_FAULT;
1273
1274 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001275 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001276
1277 /* Do not map writable pfn in the readonly memslot. */
1278 if (writable && memslot_is_readonly(slot)) {
1279 *writable = false;
1280 writable = NULL;
1281 }
1282
1283 return hva_to_pfn(addr, atomic, async, write_fault,
1284 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001285}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001286
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001287static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1288 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001289{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001290 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001291
Gleb Natapovaf585b92010-10-14 11:22:46 +02001292 if (async)
1293 *async = false;
1294
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001295 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001296
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001297 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1298 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001299}
1300
1301pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1302{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001303 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001304}
1305EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1306
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001307pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1308 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001309{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001310 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001311}
1312EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1313
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001314pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1315{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001316 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001317}
Anthony Liguori35149e22008-04-02 14:46:56 -05001318EXPORT_SYMBOL_GPL(gfn_to_pfn);
1319
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001320pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1321 bool *writable)
1322{
1323 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1324}
1325EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1326
Xiao Guangrongd5661042012-07-17 21:56:16 +08001327pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001328{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001329 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001330}
1331
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001332pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1333{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001334 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001335}
1336EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1337
Xiao Guangrong48987782010-08-22 19:11:43 +08001338int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1339 int nr_pages)
1340{
1341 unsigned long addr;
1342 gfn_t entry;
1343
Gleb Natapov49c77542010-10-18 15:22:23 +02001344 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001345 if (kvm_is_error_hva(addr))
1346 return -1;
1347
1348 if (entry < nr_pages)
1349 return 0;
1350
1351 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1352}
1353EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1354
Xiao Guangronga2766322012-07-26 11:58:59 +08001355static struct page *kvm_pfn_to_page(pfn_t pfn)
1356{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001357 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001358 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001359
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001360 if (kvm_is_mmio_pfn(pfn)) {
1361 WARN_ON(1);
1362 return KVM_ERR_PTR_BAD_PAGE;
1363 }
1364
Xiao Guangronga2766322012-07-26 11:58:59 +08001365 return pfn_to_page(pfn);
1366}
1367
Anthony Liguori35149e22008-04-02 14:46:56 -05001368struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1369{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001370 pfn_t pfn;
1371
1372 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001373
Xiao Guangronga2766322012-07-26 11:58:59 +08001374 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001375}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001376
Avi Kivity954bbbc2007-03-30 14:02:32 +03001377EXPORT_SYMBOL_GPL(gfn_to_page);
1378
Izik Eidusb4231d62007-11-20 11:49:33 +02001379void kvm_release_page_clean(struct page *page)
1380{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001381 WARN_ON(is_error_page(page));
1382
Anthony Liguori35149e22008-04-02 14:46:56 -05001383 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001384}
1385EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1386
Anthony Liguori35149e22008-04-02 14:46:56 -05001387void kvm_release_pfn_clean(pfn_t pfn)
1388{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001389 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001390 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001391}
1392EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1393
Izik Eidusb4231d62007-11-20 11:49:33 +02001394void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001395{
Xiao Guangronga2766322012-07-26 11:58:59 +08001396 WARN_ON(is_error_page(page));
1397
Anthony Liguori35149e22008-04-02 14:46:56 -05001398 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001399}
Izik Eidusb4231d62007-11-20 11:49:33 +02001400EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001401
Stephen Hemminger79408762013-12-29 12:12:29 -08001402static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001403{
1404 kvm_set_pfn_dirty(pfn);
1405 kvm_release_pfn_clean(pfn);
1406}
Anthony Liguori35149e22008-04-02 14:46:56 -05001407
1408void kvm_set_pfn_dirty(pfn_t pfn)
1409{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001410 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001411 struct page *page = pfn_to_page(pfn);
1412 if (!PageReserved(page))
1413 SetPageDirty(page);
1414 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001415}
1416EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1417
1418void kvm_set_pfn_accessed(pfn_t pfn)
1419{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001420 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001421 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001422}
1423EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1424
1425void kvm_get_pfn(pfn_t pfn)
1426{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001427 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001428 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001429}
1430EXPORT_SYMBOL_GPL(kvm_get_pfn);
1431
Izik Eidus195aefd2007-10-01 22:14:18 +02001432static int next_segment(unsigned long len, int offset)
1433{
1434 if (len > PAGE_SIZE - offset)
1435 return PAGE_SIZE - offset;
1436 else
1437 return len;
1438}
1439
1440int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1441 int len)
1442{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001443 int r;
1444 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001445
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001446 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001447 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001448 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001449 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001450 if (r)
1451 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001452 return 0;
1453}
1454EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1455
1456int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1457{
1458 gfn_t gfn = gpa >> PAGE_SHIFT;
1459 int seg;
1460 int offset = offset_in_page(gpa);
1461 int ret;
1462
1463 while ((seg = next_segment(len, offset)) != 0) {
1464 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1465 if (ret < 0)
1466 return ret;
1467 offset = 0;
1468 len -= seg;
1469 data += seg;
1470 ++gfn;
1471 }
1472 return 0;
1473}
1474EXPORT_SYMBOL_GPL(kvm_read_guest);
1475
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001476int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1477 unsigned long len)
1478{
1479 int r;
1480 unsigned long addr;
1481 gfn_t gfn = gpa >> PAGE_SHIFT;
1482 int offset = offset_in_page(gpa);
1483
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001484 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001485 if (kvm_is_error_hva(addr))
1486 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001487 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001488 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001489 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001490 if (r)
1491 return -EFAULT;
1492 return 0;
1493}
1494EXPORT_SYMBOL(kvm_read_guest_atomic);
1495
Izik Eidus195aefd2007-10-01 22:14:18 +02001496int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1497 int offset, int len)
1498{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001499 int r;
1500 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001501
Izik Eiduse0506bc2007-11-11 22:10:22 +02001502 addr = gfn_to_hva(kvm, gfn);
1503 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001504 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001505 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001506 if (r)
1507 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001508 mark_page_dirty(kvm, gfn);
1509 return 0;
1510}
1511EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1512
1513int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1514 unsigned long len)
1515{
1516 gfn_t gfn = gpa >> PAGE_SHIFT;
1517 int seg;
1518 int offset = offset_in_page(gpa);
1519 int ret;
1520
1521 while ((seg = next_segment(len, offset)) != 0) {
1522 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1523 if (ret < 0)
1524 return ret;
1525 offset = 0;
1526 len -= seg;
1527 data += seg;
1528 ++gfn;
1529 }
1530 return 0;
1531}
1532
Gleb Natapov49c77542010-10-18 15:22:23 +02001533int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001534 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001535{
1536 struct kvm_memslots *slots = kvm_memslots(kvm);
1537 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001538 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1539 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1540 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1541 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001542
1543 ghc->gpa = gpa;
1544 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001545 ghc->len = len;
1546 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1547 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1548 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001549 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001550 } else {
1551 /*
1552 * If the requested region crosses two memslots, we still
1553 * verify that the entire region is valid here.
1554 */
1555 while (start_gfn <= end_gfn) {
1556 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1557 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1558 &nr_pages_avail);
1559 if (kvm_is_error_hva(ghc->hva))
1560 return -EFAULT;
1561 start_gfn += nr_pages_avail;
1562 }
1563 /* Use the slow path for cross page reads and writes. */
1564 ghc->memslot = NULL;
1565 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001566 return 0;
1567}
1568EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1569
1570int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1571 void *data, unsigned long len)
1572{
1573 struct kvm_memslots *slots = kvm_memslots(kvm);
1574 int r;
1575
Andrew Honig8f964522013-03-29 09:35:21 -07001576 BUG_ON(len > ghc->len);
1577
Gleb Natapov49c77542010-10-18 15:22:23 +02001578 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001579 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1580
1581 if (unlikely(!ghc->memslot))
1582 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001583
1584 if (kvm_is_error_hva(ghc->hva))
1585 return -EFAULT;
1586
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001587 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001588 if (r)
1589 return -EFAULT;
1590 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1591
1592 return 0;
1593}
1594EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1595
Gleb Natapove03b6442011-07-11 15:28:11 -04001596int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1597 void *data, unsigned long len)
1598{
1599 struct kvm_memslots *slots = kvm_memslots(kvm);
1600 int r;
1601
Andrew Honig8f964522013-03-29 09:35:21 -07001602 BUG_ON(len > ghc->len);
1603
Gleb Natapove03b6442011-07-11 15:28:11 -04001604 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001605 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1606
1607 if (unlikely(!ghc->memslot))
1608 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001609
1610 if (kvm_is_error_hva(ghc->hva))
1611 return -EFAULT;
1612
1613 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1614 if (r)
1615 return -EFAULT;
1616
1617 return 0;
1618}
1619EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1620
Izik Eidus195aefd2007-10-01 22:14:18 +02001621int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1622{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001623 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1624
1625 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001626}
1627EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1628
1629int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1630{
1631 gfn_t gfn = gpa >> PAGE_SHIFT;
1632 int seg;
1633 int offset = offset_in_page(gpa);
1634 int ret;
1635
1636 while ((seg = next_segment(len, offset)) != 0) {
1637 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1638 if (ret < 0)
1639 return ret;
1640 offset = 0;
1641 len -= seg;
1642 ++gfn;
1643 }
1644 return 0;
1645}
1646EXPORT_SYMBOL_GPL(kvm_clear_guest);
1647
Stephen Hemminger79408762013-12-29 12:12:29 -08001648static void mark_page_dirty_in_slot(struct kvm *kvm,
1649 struct kvm_memory_slot *memslot,
1650 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001651{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001652 if (memslot && memslot->dirty_bitmap) {
1653 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001654
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001655 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001656 }
1657}
1658
Gleb Natapov49c77542010-10-18 15:22:23 +02001659void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1660{
1661 struct kvm_memory_slot *memslot;
1662
1663 memslot = gfn_to_memslot(kvm, gfn);
1664 mark_page_dirty_in_slot(kvm, memslot, gfn);
1665}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301666EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001667
Eddie Dongb6958ce2007-07-18 12:15:21 +03001668/*
1669 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1670 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001671void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001672{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001673 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001674
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001675 for (;;) {
1676 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001677
Gleb Natapova1b37102009-07-09 15:33:52 +03001678 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001679 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001680 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001681 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001682 if (kvm_cpu_has_pending_timer(vcpu))
1683 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001684 if (signal_pending(current))
1685 break;
1686
Eddie Dongb6958ce2007-07-18 12:15:21 +03001687 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001688 }
1689
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001690 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001691}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301692EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001693
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001694#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001695/*
1696 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1697 */
1698void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1699{
1700 int me;
1701 int cpu = vcpu->cpu;
1702 wait_queue_head_t *wqp;
1703
1704 wqp = kvm_arch_vcpu_wq(vcpu);
1705 if (waitqueue_active(wqp)) {
1706 wake_up_interruptible(wqp);
1707 ++vcpu->stat.halt_wakeup;
1708 }
1709
1710 me = get_cpu();
1711 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1712 if (kvm_arch_vcpu_should_kick(vcpu))
1713 smp_send_reschedule(cpu);
1714 put_cpu();
1715}
Yang Zhanga20ed542013-04-11 19:25:15 +08001716EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001717#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001718
Dan Carpenterfa933842014-05-23 13:20:42 +03001719int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02001720{
1721 struct pid *pid;
1722 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03001723 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001724
1725 rcu_read_lock();
1726 pid = rcu_dereference(target->pid);
1727 if (pid)
1728 task = get_pid_task(target->pid, PIDTYPE_PID);
1729 rcu_read_unlock();
1730 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301731 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001732 if (task->flags & PF_VCPU) {
1733 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301734 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001735 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301736 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001737 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301738
1739 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001740}
1741EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1742
Raghavendra K T06e48c52012-07-19 15:17:52 +05301743/*
1744 * Helper that checks whether a VCPU is eligible for directed yield.
1745 * Most eligible candidate to yield is decided by following heuristics:
1746 *
1747 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1748 * (preempted lock holder), indicated by @in_spin_loop.
1749 * Set at the beiginning and cleared at the end of interception/PLE handler.
1750 *
1751 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1752 * chance last time (mostly it has become eligible now since we have probably
1753 * yielded to lockholder in last iteration. This is done by toggling
1754 * @dy_eligible each time a VCPU checked for eligibility.)
1755 *
1756 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1757 * to preempted lock-holder could result in wrong VCPU selection and CPU
1758 * burning. Giving priority for a potential lock-holder increases lock
1759 * progress.
1760 *
1761 * Since algorithm is based on heuristics, accessing another VCPU data without
1762 * locking does not harm. It may result in trying to yield to same VCPU, fail
1763 * and continue with next VCPU and so on.
1764 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001765static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301766{
Scott Wood4a55dd72014-01-09 18:43:16 -06001767#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301768 bool eligible;
1769
1770 eligible = !vcpu->spin_loop.in_spin_loop ||
1771 (vcpu->spin_loop.in_spin_loop &&
1772 vcpu->spin_loop.dy_eligible);
1773
1774 if (vcpu->spin_loop.in_spin_loop)
1775 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1776
1777 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001778#else
1779 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301780#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001781}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301782
Rik van Riel217ece62011-02-01 09:53:28 -05001783void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001784{
Rik van Riel217ece62011-02-01 09:53:28 -05001785 struct kvm *kvm = me->kvm;
1786 struct kvm_vcpu *vcpu;
1787 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1788 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301789 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001790 int pass;
1791 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001792
Raghavendra K T4c088492012-07-18 19:07:46 +05301793 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001794 /*
1795 * We boost the priority of a VCPU that is runnable but not
1796 * currently running, because it got preempted by something
1797 * else and called schedule in __vcpu_run. Hopefully that
1798 * VCPU is holding the lock that we need and will release it.
1799 * We approximate round-robin by starting at the last boosted VCPU.
1800 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301801 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001802 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001803 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001804 i = last_boosted_vcpu;
1805 continue;
1806 } else if (pass && i > last_boosted_vcpu)
1807 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301808 if (!ACCESS_ONCE(vcpu->preempted))
1809 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001810 if (vcpu == me)
1811 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001812 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001813 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301814 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1815 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301816
1817 yielded = kvm_vcpu_yield_to(vcpu);
1818 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001819 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001820 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301821 } else if (yielded < 0) {
1822 try--;
1823 if (!try)
1824 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001825 }
Rik van Riel217ece62011-02-01 09:53:28 -05001826 }
1827 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301828 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301829
1830 /* Ensure vcpu is not eligible during next spinloop */
1831 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001832}
1833EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1834
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001835static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001836{
1837 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001838 struct page *page;
1839
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001840 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001841 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001842#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001843 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001844 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001845#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001846#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1847 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1848 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1849#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001850 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001851 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001852 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001853 vmf->page = page;
1854 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001855}
1856
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001857static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001858 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001859};
1860
1861static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1862{
1863 vma->vm_ops = &kvm_vcpu_vm_ops;
1864 return 0;
1865}
1866
Avi Kivitybccf2152007-02-21 18:04:26 +02001867static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1868{
1869 struct kvm_vcpu *vcpu = filp->private_data;
1870
Al Viro66c0b392008-04-19 20:33:56 +01001871 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001872 return 0;
1873}
1874
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001875static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001876 .release = kvm_vcpu_release,
1877 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001878#ifdef CONFIG_COMPAT
1879 .compat_ioctl = kvm_vcpu_compat_ioctl,
1880#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001881 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001882 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001883};
1884
1885/*
1886 * Allocates an inode for the vcpu.
1887 */
1888static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1889{
Yann Droneaud24009b02013-08-24 22:14:07 +02001890 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02001891}
1892
Avi Kivityc5ea7662007-02-20 18:41:05 +02001893/*
1894 * Creates some virtual cpus. Good luck creating more than one.
1895 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001896static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001897{
1898 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001899 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001900
Andy Honig338c7db2013-11-18 16:09:22 -08001901 if (id >= KVM_MAX_VCPUS)
1902 return -EINVAL;
1903
Gleb Natapov73880c82009-06-09 15:56:28 +03001904 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001905 if (IS_ERR(vcpu))
1906 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001907
Avi Kivity15ad7142007-07-11 18:17:21 +03001908 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1909
Avi Kivity26e52152007-11-20 15:30:24 +02001910 r = kvm_arch_vcpu_setup(vcpu);
1911 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001912 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001913
Shaohua Li11ec2802007-07-23 14:51:37 +08001914 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001915 if (!kvm_vcpu_compatible(vcpu)) {
1916 r = -EINVAL;
1917 goto unlock_vcpu_destroy;
1918 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001919 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1920 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001921 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001922 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001923
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001924 kvm_for_each_vcpu(r, v, kvm)
1925 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001926 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001927 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001928 }
1929
1930 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001931
1932 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001933 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001934 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001935 if (r < 0) {
1936 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001937 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001938 }
1939
1940 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1941 smp_wmb();
1942 atomic_inc(&kvm->online_vcpus);
1943
Gleb Natapov73880c82009-06-09 15:56:28 +03001944 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001945 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001946 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001947
Jan Kiszkad7805922011-05-23 10:33:05 +02001948unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001949 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001950vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001951 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001952 return r;
1953}
1954
Avi Kivity1961d272007-03-05 19:46:05 +02001955static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1956{
1957 if (sigset) {
1958 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1959 vcpu->sigset_active = 1;
1960 vcpu->sigset = *sigset;
1961 } else
1962 vcpu->sigset_active = 0;
1963 return 0;
1964}
1965
Avi Kivitybccf2152007-02-21 18:04:26 +02001966static long kvm_vcpu_ioctl(struct file *filp,
1967 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001968{
Avi Kivitybccf2152007-02-21 18:04:26 +02001969 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001970 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001971 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001972 struct kvm_fpu *fpu = NULL;
1973 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001974
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001975 if (vcpu->kvm->mm != current->mm)
1976 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001977
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001978#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001979 /*
1980 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1981 * so vcpu_load() would break it.
1982 */
1983 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1984 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1985#endif
1986
1987
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001988 r = vcpu_load(vcpu);
1989 if (r)
1990 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001991 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001992 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001993 r = -EINVAL;
1994 if (arg)
1995 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001996 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001997 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001998 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001999 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002000 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002001
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002002 r = -ENOMEM;
2003 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2004 if (!kvm_regs)
2005 goto out;
2006 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002007 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002008 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002009 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002010 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2011 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002012 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002013out_free1:
2014 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002015 break;
2016 }
2017 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002018 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002019
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002020 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002021 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2022 if (IS_ERR(kvm_regs)) {
2023 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002024 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002025 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002026 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002027 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002028 break;
2029 }
2030 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002031 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2032 r = -ENOMEM;
2033 if (!kvm_sregs)
2034 goto out;
2035 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002036 if (r)
2037 goto out;
2038 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002039 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002040 goto out;
2041 r = 0;
2042 break;
2043 }
2044 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002045 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2046 if (IS_ERR(kvm_sregs)) {
2047 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002048 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002049 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002050 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002051 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002052 break;
2053 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002054 case KVM_GET_MP_STATE: {
2055 struct kvm_mp_state mp_state;
2056
2057 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2058 if (r)
2059 goto out;
2060 r = -EFAULT;
2061 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2062 goto out;
2063 r = 0;
2064 break;
2065 }
2066 case KVM_SET_MP_STATE: {
2067 struct kvm_mp_state mp_state;
2068
2069 r = -EFAULT;
2070 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2071 goto out;
2072 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002073 break;
2074 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002075 case KVM_TRANSLATE: {
2076 struct kvm_translation tr;
2077
2078 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002079 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002080 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002081 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002082 if (r)
2083 goto out;
2084 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002085 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002086 goto out;
2087 r = 0;
2088 break;
2089 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002090 case KVM_SET_GUEST_DEBUG: {
2091 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002092
2093 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002094 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002095 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002096 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002097 break;
2098 }
Avi Kivity1961d272007-03-05 19:46:05 +02002099 case KVM_SET_SIGNAL_MASK: {
2100 struct kvm_signal_mask __user *sigmask_arg = argp;
2101 struct kvm_signal_mask kvm_sigmask;
2102 sigset_t sigset, *p;
2103
2104 p = NULL;
2105 if (argp) {
2106 r = -EFAULT;
2107 if (copy_from_user(&kvm_sigmask, argp,
2108 sizeof kvm_sigmask))
2109 goto out;
2110 r = -EINVAL;
2111 if (kvm_sigmask.len != sizeof sigset)
2112 goto out;
2113 r = -EFAULT;
2114 if (copy_from_user(&sigset, sigmask_arg->sigset,
2115 sizeof sigset))
2116 goto out;
2117 p = &sigset;
2118 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002119 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002120 break;
2121 }
Avi Kivityb8836732007-04-01 16:34:31 +03002122 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002123 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2124 r = -ENOMEM;
2125 if (!fpu)
2126 goto out;
2127 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002128 if (r)
2129 goto out;
2130 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002131 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002132 goto out;
2133 r = 0;
2134 break;
2135 }
2136 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002137 fpu = memdup_user(argp, sizeof(*fpu));
2138 if (IS_ERR(fpu)) {
2139 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002140 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002141 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002142 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002143 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002144 break;
2145 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002146 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002147 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002148 }
2149out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002150 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002151 kfree(fpu);
2152 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002153 return r;
2154}
2155
Alexander Graf1dda6062011-06-08 02:45:37 +02002156#ifdef CONFIG_COMPAT
2157static long kvm_vcpu_compat_ioctl(struct file *filp,
2158 unsigned int ioctl, unsigned long arg)
2159{
2160 struct kvm_vcpu *vcpu = filp->private_data;
2161 void __user *argp = compat_ptr(arg);
2162 int r;
2163
2164 if (vcpu->kvm->mm != current->mm)
2165 return -EIO;
2166
2167 switch (ioctl) {
2168 case KVM_SET_SIGNAL_MASK: {
2169 struct kvm_signal_mask __user *sigmask_arg = argp;
2170 struct kvm_signal_mask kvm_sigmask;
2171 compat_sigset_t csigset;
2172 sigset_t sigset;
2173
2174 if (argp) {
2175 r = -EFAULT;
2176 if (copy_from_user(&kvm_sigmask, argp,
2177 sizeof kvm_sigmask))
2178 goto out;
2179 r = -EINVAL;
2180 if (kvm_sigmask.len != sizeof csigset)
2181 goto out;
2182 r = -EFAULT;
2183 if (copy_from_user(&csigset, sigmask_arg->sigset,
2184 sizeof csigset))
2185 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002186 sigset_from_compat(&sigset, &csigset);
2187 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2188 } else
2189 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002190 break;
2191 }
2192 default:
2193 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2194 }
2195
2196out:
2197 return r;
2198}
2199#endif
2200
Scott Wood852b6d52013-04-12 14:08:42 +00002201static int kvm_device_ioctl_attr(struct kvm_device *dev,
2202 int (*accessor)(struct kvm_device *dev,
2203 struct kvm_device_attr *attr),
2204 unsigned long arg)
2205{
2206 struct kvm_device_attr attr;
2207
2208 if (!accessor)
2209 return -EPERM;
2210
2211 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2212 return -EFAULT;
2213
2214 return accessor(dev, &attr);
2215}
2216
2217static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2218 unsigned long arg)
2219{
2220 struct kvm_device *dev = filp->private_data;
2221
2222 switch (ioctl) {
2223 case KVM_SET_DEVICE_ATTR:
2224 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2225 case KVM_GET_DEVICE_ATTR:
2226 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2227 case KVM_HAS_DEVICE_ATTR:
2228 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2229 default:
2230 if (dev->ops->ioctl)
2231 return dev->ops->ioctl(dev, ioctl, arg);
2232
2233 return -ENOTTY;
2234 }
2235}
2236
Scott Wood852b6d52013-04-12 14:08:42 +00002237static int kvm_device_release(struct inode *inode, struct file *filp)
2238{
2239 struct kvm_device *dev = filp->private_data;
2240 struct kvm *kvm = dev->kvm;
2241
Scott Wood852b6d52013-04-12 14:08:42 +00002242 kvm_put_kvm(kvm);
2243 return 0;
2244}
2245
2246static const struct file_operations kvm_device_fops = {
2247 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002248#ifdef CONFIG_COMPAT
2249 .compat_ioctl = kvm_device_ioctl,
2250#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002251 .release = kvm_device_release,
2252};
2253
2254struct kvm_device *kvm_device_from_filp(struct file *filp)
2255{
2256 if (filp->f_op != &kvm_device_fops)
2257 return NULL;
2258
2259 return filp->private_data;
2260}
2261
2262static int kvm_ioctl_create_device(struct kvm *kvm,
2263 struct kvm_create_device *cd)
2264{
2265 struct kvm_device_ops *ops = NULL;
2266 struct kvm_device *dev;
2267 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2268 int ret;
2269
2270 switch (cd->type) {
Scott Wood5df554ad2013-04-12 14:08:46 +00002271#ifdef CONFIG_KVM_MPIC
2272 case KVM_DEV_TYPE_FSL_MPIC_20:
2273 case KVM_DEV_TYPE_FSL_MPIC_42:
2274 ops = &kvm_mpic_ops;
2275 break;
2276#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002277#ifdef CONFIG_KVM_XICS
2278 case KVM_DEV_TYPE_XICS:
2279 ops = &kvm_xics_ops;
2280 break;
2281#endif
Alex Williamsonec535002013-10-30 11:02:17 -06002282#ifdef CONFIG_KVM_VFIO
2283 case KVM_DEV_TYPE_VFIO:
2284 ops = &kvm_vfio_ops;
2285 break;
2286#endif
Christoffer Dall73306722013-10-25 17:29:18 +01002287#ifdef CONFIG_KVM_ARM_VGIC
2288 case KVM_DEV_TYPE_ARM_VGIC_V2:
2289 ops = &kvm_arm_vgic_v2_ops;
2290 break;
2291#endif
Jens Freimannc05c4182013-10-07 16:13:45 +02002292#ifdef CONFIG_S390
2293 case KVM_DEV_TYPE_FLIC:
2294 ops = &kvm_flic_ops;
2295 break;
2296#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002297 default:
2298 return -ENODEV;
2299 }
2300
2301 if (test)
2302 return 0;
2303
2304 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2305 if (!dev)
2306 return -ENOMEM;
2307
2308 dev->ops = ops;
2309 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002310
2311 ret = ops->create(dev, cd->type);
2312 if (ret < 0) {
2313 kfree(dev);
2314 return ret;
2315 }
2316
Yann Droneaud24009b02013-08-24 22:14:07 +02002317 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002318 if (ret < 0) {
2319 ops->destroy(dev);
2320 return ret;
2321 }
2322
Scott Wood07f0a7b2013-04-25 14:11:23 +00002323 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002324 kvm_get_kvm(kvm);
2325 cd->fd = ret;
2326 return 0;
2327}
2328
Alexander Graf92b591a2014-07-14 18:33:08 +02002329static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2330{
2331 switch (arg) {
2332 case KVM_CAP_USER_MEMORY:
2333 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2334 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2335#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2336 case KVM_CAP_SET_BOOT_CPU_ID:
2337#endif
2338 case KVM_CAP_INTERNAL_ERROR_DATA:
2339#ifdef CONFIG_HAVE_KVM_MSI
2340 case KVM_CAP_SIGNAL_MSI:
2341#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002342#ifdef CONFIG_HAVE_KVM_IRQFD
Alexander Graf92b591a2014-07-14 18:33:08 +02002343 case KVM_CAP_IRQFD_RESAMPLE:
2344#endif
2345 case KVM_CAP_CHECK_EXTENSION_VM:
2346 return 1;
2347#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2348 case KVM_CAP_IRQ_ROUTING:
2349 return KVM_MAX_IRQ_ROUTES;
2350#endif
2351 default:
2352 break;
2353 }
2354 return kvm_vm_ioctl_check_extension(kvm, arg);
2355}
2356
Avi Kivitybccf2152007-02-21 18:04:26 +02002357static long kvm_vm_ioctl(struct file *filp,
2358 unsigned int ioctl, unsigned long arg)
2359{
2360 struct kvm *kvm = filp->private_data;
2361 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002362 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002363
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002364 if (kvm->mm != current->mm)
2365 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002366 switch (ioctl) {
2367 case KVM_CREATE_VCPU:
2368 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002369 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002370 case KVM_SET_USER_MEMORY_REGION: {
2371 struct kvm_userspace_memory_region kvm_userspace_mem;
2372
2373 r = -EFAULT;
2374 if (copy_from_user(&kvm_userspace_mem, argp,
2375 sizeof kvm_userspace_mem))
2376 goto out;
2377
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002378 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002379 break;
2380 }
2381 case KVM_GET_DIRTY_LOG: {
2382 struct kvm_dirty_log log;
2383
2384 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002385 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002386 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002387 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002388 break;
2389 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002390#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2391 case KVM_REGISTER_COALESCED_MMIO: {
2392 struct kvm_coalesced_mmio_zone zone;
2393 r = -EFAULT;
2394 if (copy_from_user(&zone, argp, sizeof zone))
2395 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002396 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002397 break;
2398 }
2399 case KVM_UNREGISTER_COALESCED_MMIO: {
2400 struct kvm_coalesced_mmio_zone zone;
2401 r = -EFAULT;
2402 if (copy_from_user(&zone, argp, sizeof zone))
2403 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002404 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002405 break;
2406 }
2407#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002408 case KVM_IRQFD: {
2409 struct kvm_irqfd data;
2410
2411 r = -EFAULT;
2412 if (copy_from_user(&data, argp, sizeof data))
2413 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002414 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002415 break;
2416 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002417 case KVM_IOEVENTFD: {
2418 struct kvm_ioeventfd data;
2419
2420 r = -EFAULT;
2421 if (copy_from_user(&data, argp, sizeof data))
2422 goto out;
2423 r = kvm_ioeventfd(kvm, &data);
2424 break;
2425 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002426#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2427 case KVM_SET_BOOT_CPU_ID:
2428 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002429 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002430 if (atomic_read(&kvm->online_vcpus) != 0)
2431 r = -EBUSY;
2432 else
2433 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002434 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002435 break;
2436#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002437#ifdef CONFIG_HAVE_KVM_MSI
2438 case KVM_SIGNAL_MSI: {
2439 struct kvm_msi msi;
2440
2441 r = -EFAULT;
2442 if (copy_from_user(&msi, argp, sizeof msi))
2443 goto out;
2444 r = kvm_send_userspace_msi(kvm, &msi);
2445 break;
2446 }
2447#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002448#ifdef __KVM_HAVE_IRQ_LINE
2449 case KVM_IRQ_LINE_STATUS:
2450 case KVM_IRQ_LINE: {
2451 struct kvm_irq_level irq_event;
2452
2453 r = -EFAULT;
2454 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2455 goto out;
2456
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002457 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2458 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002459 if (r)
2460 goto out;
2461
2462 r = -EFAULT;
2463 if (ioctl == KVM_IRQ_LINE_STATUS) {
2464 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2465 goto out;
2466 }
2467
2468 r = 0;
2469 break;
2470 }
2471#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002472#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2473 case KVM_SET_GSI_ROUTING: {
2474 struct kvm_irq_routing routing;
2475 struct kvm_irq_routing __user *urouting;
2476 struct kvm_irq_routing_entry *entries;
2477
2478 r = -EFAULT;
2479 if (copy_from_user(&routing, argp, sizeof(routing)))
2480 goto out;
2481 r = -EINVAL;
2482 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2483 goto out;
2484 if (routing.flags)
2485 goto out;
2486 r = -ENOMEM;
2487 entries = vmalloc(routing.nr * sizeof(*entries));
2488 if (!entries)
2489 goto out;
2490 r = -EFAULT;
2491 urouting = argp;
2492 if (copy_from_user(entries, urouting->entries,
2493 routing.nr * sizeof(*entries)))
2494 goto out_free_irq_routing;
2495 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2496 routing.flags);
2497 out_free_irq_routing:
2498 vfree(entries);
2499 break;
2500 }
2501#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002502 case KVM_CREATE_DEVICE: {
2503 struct kvm_create_device cd;
2504
2505 r = -EFAULT;
2506 if (copy_from_user(&cd, argp, sizeof(cd)))
2507 goto out;
2508
2509 r = kvm_ioctl_create_device(kvm, &cd);
2510 if (r)
2511 goto out;
2512
2513 r = -EFAULT;
2514 if (copy_to_user(argp, &cd, sizeof(cd)))
2515 goto out;
2516
2517 r = 0;
2518 break;
2519 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002520 case KVM_CHECK_EXTENSION:
2521 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2522 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002523 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002524 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002525 if (r == -ENOTTY)
2526 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002527 }
2528out:
2529 return r;
2530}
2531
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002532#ifdef CONFIG_COMPAT
2533struct compat_kvm_dirty_log {
2534 __u32 slot;
2535 __u32 padding1;
2536 union {
2537 compat_uptr_t dirty_bitmap; /* one bit per page */
2538 __u64 padding2;
2539 };
2540};
2541
2542static long kvm_vm_compat_ioctl(struct file *filp,
2543 unsigned int ioctl, unsigned long arg)
2544{
2545 struct kvm *kvm = filp->private_data;
2546 int r;
2547
2548 if (kvm->mm != current->mm)
2549 return -EIO;
2550 switch (ioctl) {
2551 case KVM_GET_DIRTY_LOG: {
2552 struct compat_kvm_dirty_log compat_log;
2553 struct kvm_dirty_log log;
2554
2555 r = -EFAULT;
2556 if (copy_from_user(&compat_log, (void __user *)arg,
2557 sizeof(compat_log)))
2558 goto out;
2559 log.slot = compat_log.slot;
2560 log.padding1 = compat_log.padding1;
2561 log.padding2 = compat_log.padding2;
2562 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2563
2564 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002565 break;
2566 }
2567 default:
2568 r = kvm_vm_ioctl(filp, ioctl, arg);
2569 }
2570
2571out:
2572 return r;
2573}
2574#endif
2575
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002576static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002577 .release = kvm_vm_release,
2578 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002579#ifdef CONFIG_COMPAT
2580 .compat_ioctl = kvm_vm_compat_ioctl,
2581#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002582 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002583};
2584
Carsten Ottee08b9632012-01-04 10:25:20 +01002585static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002586{
Heiko Carstensaac87632010-10-27 17:22:10 +02002587 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002588 struct kvm *kvm;
2589
Carsten Ottee08b9632012-01-04 10:25:20 +01002590 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002591 if (IS_ERR(kvm))
2592 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002593#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2594 r = kvm_coalesced_mmio_init(kvm);
2595 if (r < 0) {
2596 kvm_put_kvm(kvm);
2597 return r;
2598 }
2599#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002600 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002601 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002602 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002603
Heiko Carstensaac87632010-10-27 17:22:10 +02002604 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002605}
2606
2607static long kvm_dev_ioctl(struct file *filp,
2608 unsigned int ioctl, unsigned long arg)
2609{
Avi Kivity07c45a32007-03-07 13:05:38 +02002610 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002611
2612 switch (ioctl) {
2613 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002614 r = -EINVAL;
2615 if (arg)
2616 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002617 r = KVM_API_VERSION;
2618 break;
2619 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002620 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002621 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002622 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02002623 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002624 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002625 case KVM_GET_VCPU_MMAP_SIZE:
2626 r = -EINVAL;
2627 if (arg)
2628 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002629 r = PAGE_SIZE; /* struct kvm_run */
2630#ifdef CONFIG_X86
2631 r += PAGE_SIZE; /* pio data page */
2632#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002633#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2634 r += PAGE_SIZE; /* coalesced mmio ring page */
2635#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002636 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002637 case KVM_TRACE_ENABLE:
2638 case KVM_TRACE_PAUSE:
2639 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002640 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002641 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002642 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002643 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002644 }
2645out:
2646 return r;
2647}
2648
Avi Kivity6aa8b732006-12-10 02:21:36 -08002649static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002650 .unlocked_ioctl = kvm_dev_ioctl,
2651 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002652 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002653};
2654
2655static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002656 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002657 "kvm",
2658 &kvm_chardev_ops,
2659};
2660
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002661static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002662{
2663 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002664 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002665
Rusty Russell7f59f492008-12-07 21:25:45 +10302666 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002667 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002668
Rusty Russell7f59f492008-12-07 21:25:45 +10302669 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002670
2671 r = kvm_arch_hardware_enable(NULL);
2672
2673 if (r) {
2674 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2675 atomic_inc(&hardware_enable_failed);
2676 printk(KERN_INFO "kvm: enabling virtualization on "
2677 "CPU%d failed\n", cpu);
2678 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002679}
2680
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002681static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002682{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002683 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002684 if (kvm_usage_count)
2685 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002686 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002687}
2688
2689static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002690{
2691 int cpu = raw_smp_processor_id();
2692
Rusty Russell7f59f492008-12-07 21:25:45 +10302693 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002694 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302695 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002696 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002697}
2698
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002699static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002700{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002701 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002702 if (kvm_usage_count)
2703 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002704 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002705}
2706
Alexander Graf10474ae2009-09-15 11:37:46 +02002707static void hardware_disable_all_nolock(void)
2708{
2709 BUG_ON(!kvm_usage_count);
2710
2711 kvm_usage_count--;
2712 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002713 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002714}
2715
2716static void hardware_disable_all(void)
2717{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002718 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002719 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002720 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002721}
2722
2723static int hardware_enable_all(void)
2724{
2725 int r = 0;
2726
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002727 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002728
2729 kvm_usage_count++;
2730 if (kvm_usage_count == 1) {
2731 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002732 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002733
2734 if (atomic_read(&hardware_enable_failed)) {
2735 hardware_disable_all_nolock();
2736 r = -EBUSY;
2737 }
2738 }
2739
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002740 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002741
2742 return r;
2743}
2744
Avi Kivity774c47f2007-02-12 00:54:47 -08002745static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2746 void *v)
2747{
2748 int cpu = (long)v;
2749
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002750 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002751 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002752 case CPU_DYING:
Avi Kivity6ec8a852007-08-19 15:57:26 +03002753 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2754 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002755 hardware_disable();
Avi Kivity6ec8a852007-08-19 15:57:26 +03002756 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002757 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002758 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2759 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002760 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002761 break;
2762 }
2763 return NOTIFY_OK;
2764}
2765
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002766static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002767 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002768{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002769 /*
2770 * Some (well, at least mine) BIOSes hang on reboot if
2771 * in vmx root mode.
2772 *
2773 * And Intel TXT required VMX off for all cpu when system shutdown.
2774 */
2775 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2776 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002777 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002778 return NOTIFY_OK;
2779}
2780
2781static struct notifier_block kvm_reboot_notifier = {
2782 .notifier_call = kvm_reboot,
2783 .priority = 0,
2784};
2785
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002786static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002787{
2788 int i;
2789
2790 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002791 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002792
2793 kvm_iodevice_destructor(pos);
2794 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002795 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002796}
2797
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002798static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2799 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002800{
Sasha Levin743eeb02011-07-27 16:00:48 +03002801 if (r1->addr < r2->addr)
2802 return -1;
2803 if (r1->addr + r1->len > r2->addr + r2->len)
2804 return 1;
2805 return 0;
2806}
2807
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002808static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2809{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002810 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002811}
2812
Geoff Levand39369f72013-04-05 19:20:30 +00002813static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002814 gpa_t addr, int len)
2815{
Sasha Levin743eeb02011-07-27 16:00:48 +03002816 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2817 .addr = addr,
2818 .len = len,
2819 .dev = dev,
2820 };
2821
2822 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2823 kvm_io_bus_sort_cmp, NULL);
2824
2825 return 0;
2826}
2827
Geoff Levand39369f72013-04-05 19:20:30 +00002828static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002829 gpa_t addr, int len)
2830{
2831 struct kvm_io_range *range, key;
2832 int off;
2833
2834 key = (struct kvm_io_range) {
2835 .addr = addr,
2836 .len = len,
2837 };
2838
2839 range = bsearch(&key, bus->range, bus->dev_count,
2840 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2841 if (range == NULL)
2842 return -ENOENT;
2843
2844 off = range - bus->range;
2845
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002846 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002847 off--;
2848
2849 return off;
2850}
2851
Cornelia Huck126a5af2013-07-03 16:30:53 +02002852static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2853 struct kvm_io_range *range, const void *val)
2854{
2855 int idx;
2856
2857 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2858 if (idx < 0)
2859 return -EOPNOTSUPP;
2860
2861 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002862 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002863 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2864 range->len, val))
2865 return idx;
2866 idx++;
2867 }
2868
2869 return -EOPNOTSUPP;
2870}
2871
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002872/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002873int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002874 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002875{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002876 struct kvm_io_bus *bus;
2877 struct kvm_io_range range;
2878 int r;
2879
2880 range = (struct kvm_io_range) {
2881 .addr = addr,
2882 .len = len,
2883 };
2884
2885 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2886 r = __kvm_io_bus_write(bus, &range, val);
2887 return r < 0 ? r : 0;
2888}
2889
2890/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2891int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2892 int len, const void *val, long cookie)
2893{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002894 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002895 struct kvm_io_range range;
2896
2897 range = (struct kvm_io_range) {
2898 .addr = addr,
2899 .len = len,
2900 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002901
2902 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02002903
2904 /* First try the device referenced by cookie. */
2905 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002906 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02002907 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2908 val))
2909 return cookie;
2910
2911 /*
2912 * cookie contained garbage; fall back to search and return the
2913 * correct cookie value.
2914 */
2915 return __kvm_io_bus_write(bus, &range, val);
2916}
2917
2918static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2919 void *val)
2920{
2921 int idx;
2922
2923 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03002924 if (idx < 0)
2925 return -EOPNOTSUPP;
2926
2927 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002928 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02002929 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2930 range->len, val))
2931 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03002932 idx++;
2933 }
2934
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002935 return -EOPNOTSUPP;
2936}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03002937EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002938
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002939/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002940int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2941 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002942{
Cornelia Huck126a5af2013-07-03 16:30:53 +02002943 struct kvm_io_bus *bus;
2944 struct kvm_io_range range;
2945 int r;
2946
2947 range = (struct kvm_io_range) {
2948 .addr = addr,
2949 .len = len,
2950 };
2951
2952 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2953 r = __kvm_io_bus_read(bus, &range, val);
2954 return r < 0 ? r : 0;
2955}
2956
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002957
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002958/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002959int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2960 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002961{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002962 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002963
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002964 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08002965 /* exclude ioeventfd which is limited by maximum fd */
2966 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002967 return -ENOSPC;
2968
Amos Konga13007162012-03-09 12:17:32 +08002969 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2970 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002971 if (!new_bus)
2972 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002973 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2974 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002975 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002976 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2977 synchronize_srcu_expedited(&kvm->srcu);
2978 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002979
2980 return 0;
2981}
2982
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002983/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002984int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2985 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002986{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002987 int i, r;
2988 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002989
Sasha Levincdfca7b2011-12-04 19:36:28 +02002990 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002991 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002992 for (i = 0; i < bus->dev_count; i++)
2993 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002994 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002995 break;
2996 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002997
Amos Konga13007162012-03-09 12:17:32 +08002998 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002999 return r;
Amos Konga13007162012-03-09 12:17:32 +08003000
3001 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3002 sizeof(struct kvm_io_range)), GFP_KERNEL);
3003 if (!new_bus)
3004 return -ENOMEM;
3005
3006 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3007 new_bus->dev_count--;
3008 memcpy(new_bus->range + i, bus->range + i + 1,
3009 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003010
3011 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3012 synchronize_srcu_expedited(&kvm->srcu);
3013 kfree(bus);
3014 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003015}
3016
Avi Kivity774c47f2007-02-12 00:54:47 -08003017static struct notifier_block kvm_cpu_notifier = {
3018 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003019};
3020
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003021static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003022{
3023 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003024 struct kvm *kvm;
3025
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003026 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003027 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003028 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003029 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003030 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003031 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003032}
3033
3034DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3035
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003036static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003037{
3038 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003039 struct kvm *kvm;
3040 struct kvm_vcpu *vcpu;
3041 int i;
3042
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003043 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003044 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003045 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003046 kvm_for_each_vcpu(i, vcpu, kvm)
3047 *val += *(u32 *)((void *)vcpu + offset);
3048
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003049 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003050 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003051}
3052
Avi Kivityba1389b2007-11-18 16:24:12 +02003053DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3054
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003055static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003056 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3057 [KVM_STAT_VM] = &vm_stat_fops,
3058};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003059
Hamo4f69b682011-12-15 14:23:16 +08003060static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003061{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003062 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003063 struct kvm_stats_debugfs_item *p;
3064
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003065 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003066 if (kvm_debugfs_dir == NULL)
3067 goto out;
3068
3069 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003070 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003071 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003072 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003073 if (p->dentry == NULL)
3074 goto out_dir;
3075 }
3076
3077 return 0;
3078
3079out_dir:
3080 debugfs_remove_recursive(kvm_debugfs_dir);
3081out:
3082 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003083}
3084
3085static void kvm_exit_debug(void)
3086{
3087 struct kvm_stats_debugfs_item *p;
3088
3089 for (p = debugfs_entries; p->name; ++p)
3090 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003091 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003092}
3093
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003094static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003095{
Alexander Graf10474ae2009-09-15 11:37:46 +02003096 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003097 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003098 return 0;
3099}
3100
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003101static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003102{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003103 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003104 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003105 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003106 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003107}
3108
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003109static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003110 .suspend = kvm_suspend,
3111 .resume = kvm_resume,
3112};
3113
Avi Kivity15ad7142007-07-11 18:17:21 +03003114static inline
3115struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3116{
3117 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3118}
3119
3120static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3121{
3122 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303123 if (vcpu->preempted)
3124 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003125
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003126 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003127}
3128
3129static void kvm_sched_out(struct preempt_notifier *pn,
3130 struct task_struct *next)
3131{
3132 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3133
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303134 if (current->state == TASK_RUNNING)
3135 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003136 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003137}
3138
Avi Kivity0ee75be2010-04-28 15:39:01 +03003139int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003140 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003141{
3142 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003143 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003144
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003145 r = kvm_arch_init(opaque);
3146 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003147 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003148
Asias He7dac16c2013-05-08 10:57:29 +08003149 /*
3150 * kvm_arch_init makes sure there's at most one caller
3151 * for architectures that support multiple implementations,
3152 * like intel and amd on x86.
3153 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3154 * conflicts in case kvm is already setup for another implementation.
3155 */
3156 r = kvm_irqfd_init();
3157 if (r)
3158 goto out_irqfd;
3159
Avi Kivity8437a612009-06-06 14:52:35 -07003160 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303161 r = -ENOMEM;
3162 goto out_free_0;
3163 }
3164
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003165 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003166 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303167 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003168
Yang, Sheng002c7f72007-07-31 14:23:01 +03003169 for_each_online_cpu(cpu) {
3170 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003171 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003172 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003173 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003174 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003175 }
3176
Avi Kivity774c47f2007-02-12 00:54:47 -08003177 r = register_cpu_notifier(&kvm_cpu_notifier);
3178 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003179 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003180 register_reboot_notifier(&kvm_reboot_notifier);
3181
Rusty Russellc16f8622007-07-30 21:12:19 +10003182 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003183 if (!vcpu_align)
3184 vcpu_align = __alignof__(struct kvm_vcpu);
3185 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003186 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003187 if (!kvm_vcpu_cache) {
3188 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003189 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003190 }
3191
Gleb Natapovaf585b92010-10-14 11:22:46 +02003192 r = kvm_async_pf_init();
3193 if (r)
3194 goto out_free;
3195
Avi Kivity6aa8b732006-12-10 02:21:36 -08003196 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003197 kvm_vm_fops.owner = module;
3198 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003199
3200 r = misc_register(&kvm_dev);
3201 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003202 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003203 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003204 }
3205
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003206 register_syscore_ops(&kvm_syscore_ops);
3207
Avi Kivity15ad7142007-07-11 18:17:21 +03003208 kvm_preempt_ops.sched_in = kvm_sched_in;
3209 kvm_preempt_ops.sched_out = kvm_sched_out;
3210
Hamo4f69b682011-12-15 14:23:16 +08003211 r = kvm_init_debug();
3212 if (r) {
3213 printk(KERN_ERR "kvm: create debugfs files failed\n");
3214 goto out_undebugfs;
3215 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003216
Avi Kivityc7addb92007-09-16 18:58:32 +02003217 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003218
Hamo4f69b682011-12-15 14:23:16 +08003219out_undebugfs:
3220 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003221 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003222out_unreg:
3223 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003224out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003225 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003226out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003227 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003228 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003229out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003230out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003231 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303232out_free_0a:
3233 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003234out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003235 kvm_irqfd_exit();
3236out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003237 kvm_arch_exit();
3238out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003239 return r;
3240}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003241EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003242
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003243void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003244{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003245 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003246 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003247 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003248 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003249 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003250 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003251 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003252 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003253 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003254 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003255 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303256 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003257}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003258EXPORT_SYMBOL_GPL(kvm_exit);