blob: 1580dd4ace4eac20b37043c2f5c882349204ed4a [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
Jan Kiszkae935b832011-02-08 12:55:33 +010073DEFINE_RAW_SPINLOCK(kvm_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080075
Rusty Russell7f59f492008-12-07 21:25:45 +103076static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020077static int kvm_usage_count = 0;
78static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030079
Rusty Russellc16f8622007-07-30 21:12:19 +100080struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030082
Avi Kivity15ad7142007-07-11 18:17:21 +030083static __read_mostly struct preempt_ops kvm_preempt_ops;
84
Hollis Blanchard76f7c872008-04-15 16:05:42 -050085struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080086
Avi Kivitybccf2152007-02-21 18:04:26 +020087static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020089#ifdef CONFIG_COMPAT
90static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020093static int hardware_enable_all(void);
94static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020095
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020096static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
Avi Kivityb7c41452010-12-02 17:52:50 +020098bool kvm_rebooting;
99EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300100
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300101static bool largepages_enabled = true;
102
Xiao Guangronga2766322012-07-26 11:58:59 +0800103bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300104{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100105 if (pfn_valid(pfn)) {
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800106 int reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800107 struct page *tail = pfn_to_page(pfn);
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800110 if (head != tail) {
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800111 /*
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
119 * what we just read.
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800120 */
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800121 smp_rmb();
122 if (PageTail(tail))
123 return reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800124 }
125 return PageReserved(tail);
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100126 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300127
128 return true;
129}
130
Avi Kivity6aa8b732006-12-10 02:21:36 -0800131/*
132 * Switches to specified vcpu, until a matching vcpu_put()
133 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300134int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135{
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 int cpu;
137
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300138 if (mutex_lock_killable(&vcpu->mutex))
139 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
145 synchronize_rcu();
146 put_pid(oldpid);
147 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300148 cpu = get_cpu();
149 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200150 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300151 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300152 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200153}
154
Carsten Otte313a3dc2007-10-11 19:16:52 +0200155void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800156{
Avi Kivity15ad7142007-07-11 18:17:21 +0300157 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200158 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300159 preempt_notifier_unregister(&vcpu->preempt_notifier);
160 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800161 mutex_unlock(&vcpu->mutex);
162}
163
Avi Kivityd9e368d2007-06-07 19:18:30 +0300164static void ack_flush(void *_completed)
165{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300166}
167
Rusty Russell49846892008-12-08 20:26:24 +1030168static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300169{
Avi Kivity597a5f52008-07-20 14:24:22 +0300170 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030171 cpumask_var_t cpus;
172 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300173 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300174
Li Zefan79f55992009-06-15 14:58:26 +0800175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800177 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300178 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800179 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300180 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800181
182 /* Set ->requests bit before we read ->mode */
183 smp_mb();
184
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030187 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300188 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
193 else
194 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800195 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030196 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030197 return called;
198}
199
200void kvm_flush_remote_tlbs(struct kvm *kvm)
201{
Alex Shibec87d62012-03-04 13:13:38 +0800202 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800203
204 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300208}
209
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500210void kvm_reload_remote_mmus(struct kvm *kvm)
211{
Rusty Russell49846892008-12-08 20:26:24 +1030212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500213}
214
Marcelo Tosattid8281992012-11-27 23:29:01 -0200215void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216{
217 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218}
219
Yang Zhang3d81bc72013-04-11 19:25:13 +0800220void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800221{
Yang Zhang3d81bc72013-04-11 19:25:13 +0800222 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800223}
224
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000225int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
226{
227 struct page *page;
228 int r;
229
230 mutex_init(&vcpu->mutex);
231 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232 vcpu->kvm = kvm;
233 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500234 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300235 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200236 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000237
238 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
239 if (!page) {
240 r = -ENOMEM;
241 goto fail;
242 }
243 vcpu->run = page_address(page);
244
Raghavendra K T4c088492012-07-18 19:07:46 +0530245 kvm_vcpu_set_in_spin_loop(vcpu, false);
246 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530247 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530248
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800249 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000250 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800251 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000252 return 0;
253
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000254fail_free_run:
255 free_page((unsigned long)vcpu->run);
256fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000257 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000258}
259EXPORT_SYMBOL_GPL(kvm_vcpu_init);
260
261void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
262{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500263 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800264 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000265 free_page((unsigned long)vcpu->run);
266}
267EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
268
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200269#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
270static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
271{
272 return container_of(mn, struct kvm, mmu_notifier);
273}
274
275static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
276 struct mm_struct *mm,
277 unsigned long address)
278{
279 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200280 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200281
282 /*
283 * When ->invalidate_page runs, the linux pte has been zapped
284 * already but the page is still allocated until
285 * ->invalidate_page returns. So if we increase the sequence
286 * here the kvm page fault will notice if the spte can't be
287 * established because the page is going to be freed. If
288 * instead the kvm page fault establishes the spte before
289 * ->invalidate_page runs, kvm_unmap_hva will release it
290 * before returning.
291 *
292 * The sequence increase only need to be seen at spin_unlock
293 * time, and not at spin_lock time.
294 *
295 * Increasing the sequence after the spin_unlock would be
296 * unsafe because the kvm page fault could then establish the
297 * pte after kvm_unmap_hva returned, without noticing the page
298 * is going to be freed.
299 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200300 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200301 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900302
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200303 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800304 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200305 /* we've to flush the tlb before the pages can be freed */
306 if (need_tlb_flush)
307 kvm_flush_remote_tlbs(kvm);
308
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900309 spin_unlock(&kvm->mmu_lock);
310 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200311}
312
Izik Eidus3da0dd42009-09-23 21:47:18 +0300313static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
314 struct mm_struct *mm,
315 unsigned long address,
316 pte_t pte)
317{
318 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200319 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300320
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200321 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300322 spin_lock(&kvm->mmu_lock);
323 kvm->mmu_notifier_seq++;
324 kvm_set_spte_hva(kvm, address, pte);
325 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200326 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300327}
328
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200329static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
330 struct mm_struct *mm,
331 unsigned long start,
332 unsigned long end)
333{
334 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200335 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200336
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200337 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200338 spin_lock(&kvm->mmu_lock);
339 /*
340 * The count increase must become visible at unlock time as no
341 * spte can be established without taking the mmu_lock and
342 * count is also read inside the mmu_lock critical section.
343 */
344 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900345 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800346 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200347 /* we've to flush the tlb before the pages can be freed */
348 if (need_tlb_flush)
349 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900350
351 spin_unlock(&kvm->mmu_lock);
352 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200353}
354
355static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
356 struct mm_struct *mm,
357 unsigned long start,
358 unsigned long end)
359{
360 struct kvm *kvm = mmu_notifier_to_kvm(mn);
361
362 spin_lock(&kvm->mmu_lock);
363 /*
364 * This sequence increase will notify the kvm page fault that
365 * the page that is going to be mapped in the spte could have
366 * been freed.
367 */
368 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000369 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200370 /*
371 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000372 * below count decrease, which is ensured by the smp_wmb above
373 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200374 */
375 kvm->mmu_notifier_count--;
376 spin_unlock(&kvm->mmu_lock);
377
378 BUG_ON(kvm->mmu_notifier_count < 0);
379}
380
381static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
382 struct mm_struct *mm,
383 unsigned long address)
384{
385 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200386 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200387
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200388 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200389 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200390
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900391 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200392 if (young)
393 kvm_flush_remote_tlbs(kvm);
394
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900395 spin_unlock(&kvm->mmu_lock);
396 srcu_read_unlock(&kvm->srcu, idx);
397
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200398 return young;
399}
400
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800401static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
402 struct mm_struct *mm,
403 unsigned long address)
404{
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 int young, idx;
407
408 idx = srcu_read_lock(&kvm->srcu);
409 spin_lock(&kvm->mmu_lock);
410 young = kvm_test_age_hva(kvm, address);
411 spin_unlock(&kvm->mmu_lock);
412 srcu_read_unlock(&kvm->srcu, idx);
413
414 return young;
415}
416
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100417static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
418 struct mm_struct *mm)
419{
420 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800421 int idx;
422
423 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300424 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800425 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100426}
427
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200428static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
429 .invalidate_page = kvm_mmu_notifier_invalidate_page,
430 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
431 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
432 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800433 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300434 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100435 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200436};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200437
438static int kvm_init_mmu_notifier(struct kvm *kvm)
439{
440 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
441 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
442}
443
444#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
445
446static int kvm_init_mmu_notifier(struct kvm *kvm)
447{
448 return 0;
449}
450
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200451#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
452
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800453static void kvm_init_memslots_id(struct kvm *kvm)
454{
455 int i;
456 struct kvm_memslots *slots = kvm->memslots;
457
458 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800459 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800460}
461
Carsten Ottee08b9632012-01-04 10:25:20 +0100462static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800463{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100464 int r, i;
465 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800466
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100467 if (!kvm)
468 return ERR_PTR(-ENOMEM);
469
Carsten Ottee08b9632012-01-04 10:25:20 +0100470 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100471 if (r)
472 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200473
474 r = hardware_enable_all();
475 if (r)
476 goto out_err_nodisable;
477
Avi Kivity75858a82009-01-04 17:10:50 +0200478#ifdef CONFIG_HAVE_KVM_IRQCHIP
479 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300480 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200481#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800482
Alex Williamson1e702d92012-12-10 10:33:32 -0700483 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
484
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200485 r = -ENOMEM;
486 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
487 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100488 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800489 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200490 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100491 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200492 for (i = 0; i < KVM_NR_BUSES; i++) {
493 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
494 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100495 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200496 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200497 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700498
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900499 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200500 kvm->mm = current->mm;
501 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400502 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800503 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300504 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200505 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300506 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000507 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700508
509 r = kvm_init_mmu_notifier(kvm);
510 if (r)
511 goto out_err;
512
Jan Kiszkae935b832011-02-08 12:55:33 +0100513 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000514 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100515 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100516
Avi Kivityf17abe92007-02-21 19:28:04 +0200517 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200518
519out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100520 cleanup_srcu_struct(&kvm->srcu);
521out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200522 hardware_disable_all();
523out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200524 for (i = 0; i < KVM_NR_BUSES; i++)
525 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200526 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100527 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200528 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200529}
530
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900531/*
532 * Avoid using vmalloc for a small buffer.
533 * Should not be used when the size is statically known.
534 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900535void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900536{
537 if (size > PAGE_SIZE)
538 return vzalloc(size);
539 else
540 return kzalloc(size, GFP_KERNEL);
541}
542
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900543void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900544{
545 if (is_vmalloc_addr(addr))
546 vfree(addr);
547 else
548 kfree(addr);
549}
550
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900551static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
552{
553 if (!memslot->dirty_bitmap)
554 return;
555
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900556 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900557 memslot->dirty_bitmap = NULL;
558}
559
Avi Kivity6aa8b732006-12-10 02:21:36 -0800560/*
561 * Free any memory in @free but not in @dont.
562 */
563static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
564 struct kvm_memory_slot *dont)
565{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900567 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800568
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900569 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300570
Avi Kivity6aa8b732006-12-10 02:21:36 -0800571 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800572}
573
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800574void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800575{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200576 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800577 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800578
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800579 kvm_for_each_memslot(memslot, slots)
580 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200581
582 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800583}
584
Scott Wood07f0a7b2013-04-25 14:11:23 +0000585static void kvm_destroy_devices(struct kvm *kvm)
586{
587 struct list_head *node, *tmp;
588
589 list_for_each_safe(node, tmp, &kvm->devices) {
590 struct kvm_device *dev =
591 list_entry(node, struct kvm_device, vm_node);
592
593 list_del(node);
594 dev->ops->destroy(dev);
595 }
596}
597
Avi Kivityf17abe92007-02-21 19:28:04 +0200598static void kvm_destroy_vm(struct kvm *kvm)
599{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200600 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200601 struct mm_struct *mm = kvm->mm;
602
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800603 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100604 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800605 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100606 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200607 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200608 for (i = 0; i < KVM_NR_BUSES; i++)
609 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200610 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200611#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
612 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200613#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300614 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200615#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800616 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000617 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100618 kvm_free_physmem(kvm);
619 cleanup_srcu_struct(&kvm->srcu);
620 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200621 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200622 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200623}
624
Izik Eidusd39f13b2008-03-30 16:01:25 +0300625void kvm_get_kvm(struct kvm *kvm)
626{
627 atomic_inc(&kvm->users_count);
628}
629EXPORT_SYMBOL_GPL(kvm_get_kvm);
630
631void kvm_put_kvm(struct kvm *kvm)
632{
633 if (atomic_dec_and_test(&kvm->users_count))
634 kvm_destroy_vm(kvm);
635}
636EXPORT_SYMBOL_GPL(kvm_put_kvm);
637
638
Avi Kivityf17abe92007-02-21 19:28:04 +0200639static int kvm_vm_release(struct inode *inode, struct file *filp)
640{
641 struct kvm *kvm = filp->private_data;
642
Gregory Haskins721eecb2009-05-20 10:30:49 -0400643 kvm_irqfd_release(kvm);
644
Izik Eidusd39f13b2008-03-30 16:01:25 +0300645 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800646 return 0;
647}
648
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900649/*
650 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900651 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900652 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900653static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
654{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900655#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900656 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900657
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900658 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900659 if (!memslot->dirty_bitmap)
660 return -ENOMEM;
661
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900662#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900663 return 0;
664}
665
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800666static int cmp_memslot(const void *slot1, const void *slot2)
667{
668 struct kvm_memory_slot *s1, *s2;
669
670 s1 = (struct kvm_memory_slot *)slot1;
671 s2 = (struct kvm_memory_slot *)slot2;
672
673 if (s1->npages < s2->npages)
674 return 1;
675 if (s1->npages > s2->npages)
676 return -1;
677
678 return 0;
679}
680
681/*
682 * Sort the memslots base on its size, so the larger slots
683 * will get better fit.
684 */
685static void sort_memslots(struct kvm_memslots *slots)
686{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800687 int i;
688
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800689 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
690 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800691
692 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
693 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800694}
695
Alex Williamson116c14c2012-12-21 08:20:16 -0700696void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
697 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800698{
699 if (new) {
700 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800701 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800702 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800703
Xiao Guangrong28a37542011-11-24 19:04:35 +0800704 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800705 if (new->npages != npages)
706 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800707 }
708
Alex Williamson116c14c2012-12-21 08:20:16 -0700709 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800710}
711
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800712static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
713{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800714 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
715
716#ifdef KVM_CAP_READONLY_MEM
717 valid_flags |= KVM_MEM_READONLY;
718#endif
719
720 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800721 return -EINVAL;
722
723 return 0;
724}
725
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200726static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
727 struct kvm_memslots *slots, struct kvm_memory_slot *new)
728{
729 struct kvm_memslots *old_memslots = kvm->memslots;
730
731 update_memslots(slots, new, kvm->memslots->generation);
732 rcu_assign_pointer(kvm->memslots, slots);
733 synchronize_srcu_expedited(&kvm->srcu);
734 return old_memslots;
735}
736
Avi Kivity6aa8b732006-12-10 02:21:36 -0800737/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800738 * Allocate some memory and give it an address in the guest physical address
739 * space.
740 *
741 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800742 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500743 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800744 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800745int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900746 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800747{
Avi Kivity8234b222010-12-27 12:08:45 +0200748 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800749 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200750 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900751 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800752 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700753 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900754 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800755
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800756 r = check_memory_region_flags(mem);
757 if (r)
758 goto out;
759
Avi Kivity6aa8b732006-12-10 02:21:36 -0800760 r = -EINVAL;
761 /* General sanity checks */
762 if (mem->memory_size & (PAGE_SIZE - 1))
763 goto out;
764 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
765 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900766 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900767 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900768 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200769 !access_ok(VERIFY_WRITE,
770 (void __user *)(unsigned long)mem->userspace_addr,
771 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600772 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800773 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800774 goto out;
775 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
776 goto out;
777
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900778 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
780 npages = mem->memory_size >> PAGE_SHIFT;
781
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900782 r = -EINVAL;
783 if (npages > KVM_MEM_MAX_NR_PAGES)
784 goto out;
785
Avi Kivity6aa8b732006-12-10 02:21:36 -0800786 if (!npages)
787 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
788
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900789 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800790
Avi Kivitye36d96f2010-06-21 10:56:36 +0300791 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800792 new.base_gfn = base_gfn;
793 new.npages = npages;
794 new.flags = mem->flags;
795
Avi Kivity6aa8b732006-12-10 02:21:36 -0800796 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900797 if (npages) {
798 if (!old.npages)
799 change = KVM_MR_CREATE;
800 else { /* Modify an existing slot. */
801 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900802 (npages != old.npages) ||
803 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900804 goto out;
805
806 if (base_gfn != old.base_gfn)
807 change = KVM_MR_MOVE;
808 else if (new.flags != old.flags)
809 change = KVM_MR_FLAGS_ONLY;
810 else { /* Nothing to change. */
811 r = 0;
812 goto out;
813 }
814 }
815 } else if (old.npages) {
816 change = KVM_MR_DELETE;
817 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900818 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800819
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900820 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900821 /* Check for overlaps */
822 r = -EEXIST;
823 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900824 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
825 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900826 continue;
827 if (!((base_gfn + npages <= slot->base_gfn) ||
828 (base_gfn >= slot->base_gfn + slot->npages)))
829 goto out;
830 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800831 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800832
Avi Kivity6aa8b732006-12-10 02:21:36 -0800833 /* Free page dirty bitmap if unneeded */
834 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000835 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800836
837 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900838 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900839 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900840
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900841 if (kvm_arch_create_memslot(&new, npages))
842 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800843 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200844
Avi Kivity6aa8b732006-12-10 02:21:36 -0800845 /* Allocate page dirty bitmap if needed */
846 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900847 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800848 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800849 }
850
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900851 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200852 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100853 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
854 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200855 if (!slots)
856 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800857 slot = id_to_memslot(slots, mem->slot);
858 slot->flags |= KVM_MEMSLOT_INVALID;
859
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200860 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200861
Alex Williamsone40f1932012-12-10 10:32:57 -0700862 /* slot was deleted or moved, clear iommu mapping */
863 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300864 /* From this point no new shadow pages pointing to a deleted,
865 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200866 *
867 * validation of sp->gfn happens in:
868 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
869 * - kvm_is_visible_gfn (mmu_check_roots)
870 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300871 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700872 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200873 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300874
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900875 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200876 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700877 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200878
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200879 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700880 /*
881 * We can re-use the old_memslots from above, the only difference
882 * from the currently installed memslots is the invalid flag. This
883 * will get overwritten by update_memslots anyway.
884 */
885 if (!slots) {
886 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
887 GFP_KERNEL);
888 if (!slots)
889 goto out_free;
890 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200891
Alex Williamson261874b2013-01-24 15:04:03 -0700892 /*
893 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900894 * un-mapped and re-mapped if their base changes. Since base change
895 * unmapping is handled above with slot deletion, mapping alone is
896 * needed here. Anything else the iommu might care about for existing
897 * slots (size changes, userspace addr changes and read-only flag
898 * changes) is disallowed above, so any other attribute changes getting
899 * here can be skipped.
Alex Williamson261874b2013-01-24 15:04:03 -0700900 */
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900901 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
902 r = kvm_iommu_map_pages(kvm, &new);
903 if (r)
904 goto out_slots;
Alex Williamsone40f1932012-12-10 10:32:57 -0700905 }
906
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200907 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900908 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200909 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900910 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200911 }
912
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200913 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800914
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900915 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800916
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200917 kvm_free_physmem_slot(&old, &new);
918 kfree(old_memslots);
919
Avi Kivity6aa8b732006-12-10 02:21:36 -0800920 return 0;
921
Alex Williamsone40f1932012-12-10 10:32:57 -0700922out_slots:
923 kfree(slots);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800924out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800925 kvm_free_physmem_slot(&new, &old);
926out:
927 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200928}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800929EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
930
931int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900932 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800933{
934 int r;
935
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200936 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900937 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200938 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800939 return r;
940}
Izik Eidus210c7c42007-10-24 23:52:57 +0200941EXPORT_SYMBOL_GPL(kvm_set_memory_region);
942
Carsten Otte1fe779f2007-10-29 16:08:35 +0100943int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900944 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200945{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700946 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200947 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900948 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800949}
950
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800951int kvm_get_dirty_log(struct kvm *kvm,
952 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800953{
954 struct kvm_memory_slot *memslot;
955 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900956 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957 unsigned long any = 0;
958
Avi Kivity6aa8b732006-12-10 02:21:36 -0800959 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700960 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800961 goto out;
962
Xiao Guangrong28a37542011-11-24 19:04:35 +0800963 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800964 r = -ENOENT;
965 if (!memslot->dirty_bitmap)
966 goto out;
967
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900968 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969
Uri Lublincd1a4a92007-02-22 16:43:09 +0200970 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800971 any = memslot->dirty_bitmap[i];
972
973 r = -EFAULT;
974 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
975 goto out;
976
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800977 if (any)
978 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800979
980 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800981out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800982 return r;
983}
984
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900985bool kvm_largepages_enabled(void)
986{
987 return largepages_enabled;
988}
989
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300990void kvm_disable_largepages(void)
991{
992 largepages_enabled = false;
993}
994EXPORT_SYMBOL_GPL(kvm_disable_largepages);
995
Gleb Natapov49c77542010-10-18 15:22:23 +0200996struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
997{
998 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
999}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001000EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001001
Izik Eiduse0d62c72007-10-24 23:57:46 +02001002int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1003{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001004 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001005
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001006 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001007 memslot->flags & KVM_MEMSLOT_INVALID)
1008 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001009
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001010 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001011}
1012EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1013
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001014unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1015{
1016 struct vm_area_struct *vma;
1017 unsigned long addr, size;
1018
1019 size = PAGE_SIZE;
1020
1021 addr = gfn_to_hva(kvm, gfn);
1022 if (kvm_is_error_hva(addr))
1023 return PAGE_SIZE;
1024
1025 down_read(&current->mm->mmap_sem);
1026 vma = find_vma(current->mm, addr);
1027 if (!vma)
1028 goto out;
1029
1030 size = vma_kernel_pagesize(vma);
1031
1032out:
1033 up_read(&current->mm->mmap_sem);
1034
1035 return size;
1036}
1037
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001038static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1039{
1040 return slot->flags & KVM_MEM_READONLY;
1041}
1042
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001043static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001045{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001046 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001047 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001048
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001049 if (memslot_is_readonly(slot) && write)
1050 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001051
1052 if (nr_pages)
1053 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1054
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001056}
Xiao Guangrong48987782010-08-22 19:11:43 +08001057
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001058static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1059 gfn_t *nr_pages)
1060{
1061 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1062}
1063
1064unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1065 gfn_t gfn)
1066{
1067 return gfn_to_hva_many(slot, gfn, NULL);
1068}
1069EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1070
Xiao Guangrong48987782010-08-22 19:11:43 +08001071unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1072{
Gleb Natapov49c77542010-10-18 15:22:23 +02001073 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001074}
Sheng Yang0d150292008-04-25 21:44:50 +08001075EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001076
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001077/*
1078 * The hva returned by this function is only allowed to be read.
1079 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1080 */
1081static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001082{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001083 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001084}
1085
1086static int kvm_read_hva(void *data, void __user *hva, int len)
1087{
1088 return __copy_from_user(data, hva, len);
1089}
1090
1091static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1092{
1093 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001094}
1095
Geoff Levand39369f72013-04-05 19:20:30 +00001096static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001097 unsigned long start, int write, struct page **page)
1098{
1099 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1100
1101 if (write)
1102 flags |= FOLL_WRITE;
1103
1104 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1105}
1106
Huang Yingfafc3db2011-01-30 11:15:49 +08001107static inline int check_user_page_hwpoison(unsigned long addr)
1108{
1109 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1110
1111 rc = __get_user_pages(current, current->mm, addr, 1,
1112 flags, NULL, NULL, NULL);
1113 return rc == -EHWPOISON;
1114}
1115
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001116/*
1117 * The atomic path to get the writable pfn which will be stored in @pfn,
1118 * true indicates success, otherwise false is returned.
1119 */
1120static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1121 bool write_fault, bool *writable, pfn_t *pfn)
1122{
1123 struct page *page[1];
1124 int npages;
1125
1126 if (!(async || atomic))
1127 return false;
1128
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001129 /*
1130 * Fast pin a writable pfn only if it is a write fault request
1131 * or the caller allows to map a writable pfn for a read fault
1132 * request.
1133 */
1134 if (!(write_fault || writable))
1135 return false;
1136
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001137 npages = __get_user_pages_fast(addr, 1, 1, page);
1138 if (npages == 1) {
1139 *pfn = page_to_pfn(page[0]);
1140
1141 if (writable)
1142 *writable = true;
1143 return true;
1144 }
1145
1146 return false;
1147}
1148
1149/*
1150 * The slow path to get the pfn of the specified host virtual address,
1151 * 1 indicates success, -errno is returned if error is detected.
1152 */
1153static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1154 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001155{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001156 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001157 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001158
1159 might_sleep();
1160
1161 if (writable)
1162 *writable = write_fault;
1163
1164 if (async) {
1165 down_read(&current->mm->mmap_sem);
1166 npages = get_user_page_nowait(current, current->mm,
1167 addr, write_fault, page);
1168 up_read(&current->mm->mmap_sem);
1169 } else
1170 npages = get_user_pages_fast(addr, 1, write_fault,
1171 page);
1172 if (npages != 1)
1173 return npages;
1174
1175 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001176 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001177 struct page *wpage[1];
1178
1179 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1180 if (npages == 1) {
1181 *writable = true;
1182 put_page(page[0]);
1183 page[0] = wpage[0];
1184 }
1185
1186 npages = 1;
1187 }
1188 *pfn = page_to_pfn(page[0]);
1189 return npages;
1190}
1191
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001192static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1193{
1194 if (unlikely(!(vma->vm_flags & VM_READ)))
1195 return false;
1196
1197 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1198 return false;
1199
1200 return true;
1201}
1202
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001203/*
1204 * Pin guest page in memory and return its pfn.
1205 * @addr: host virtual address which maps memory to the guest
1206 * @atomic: whether this function can sleep
1207 * @async: whether this function need to wait IO complete if the
1208 * host page is not in the memory
1209 * @write_fault: whether we should get a writable host page
1210 * @writable: whether it allows to map a writable host page for !@write_fault
1211 *
1212 * The function will map a writable host page for these two cases:
1213 * 1): @write_fault = true
1214 * 2): @write_fault = false && @writable, @writable will tell the caller
1215 * whether the mapping is writable.
1216 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001217static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1218 bool write_fault, bool *writable)
1219{
1220 struct vm_area_struct *vma;
1221 pfn_t pfn = 0;
1222 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001223
Gleb Natapovaf585b92010-10-14 11:22:46 +02001224 /* we can do it either atomically or asynchronously, not both */
1225 BUG_ON(atomic && async);
1226
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001227 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1228 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001229
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001230 if (atomic)
1231 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001232
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001233 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1234 if (npages == 1)
1235 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001236
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001237 down_read(&current->mm->mmap_sem);
1238 if (npages == -EHWPOISON ||
1239 (!async && check_user_page_hwpoison(addr))) {
1240 pfn = KVM_PFN_ERR_HWPOISON;
1241 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001242 }
Izik Eidus539cb662007-11-11 22:05:04 +02001243
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001244 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001245
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001246 if (vma == NULL)
1247 pfn = KVM_PFN_ERR_FAULT;
1248 else if ((vma->vm_flags & VM_PFNMAP)) {
1249 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1250 vma->vm_pgoff;
1251 BUG_ON(!kvm_is_mmio_pfn(pfn));
1252 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001253 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001254 *async = true;
1255 pfn = KVM_PFN_ERR_FAULT;
1256 }
1257exit:
1258 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001259 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001260}
1261
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001262static pfn_t
1263__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1264 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001265{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001266 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1267
1268 if (addr == KVM_HVA_ERR_RO_BAD)
1269 return KVM_PFN_ERR_RO_FAULT;
1270
1271 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001272 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001273
1274 /* Do not map writable pfn in the readonly memslot. */
1275 if (writable && memslot_is_readonly(slot)) {
1276 *writable = false;
1277 writable = NULL;
1278 }
1279
1280 return hva_to_pfn(addr, atomic, async, write_fault,
1281 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001282}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001283
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001284static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1285 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001286{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001287 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001288
Gleb Natapovaf585b92010-10-14 11:22:46 +02001289 if (async)
1290 *async = false;
1291
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001292 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001293
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001294 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1295 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001296}
1297
1298pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1299{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001300 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001301}
1302EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1303
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001304pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1305 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001306{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001307 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001308}
1309EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1310
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001311pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1312{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001313 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001314}
Anthony Liguori35149e22008-04-02 14:46:56 -05001315EXPORT_SYMBOL_GPL(gfn_to_pfn);
1316
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001317pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1318 bool *writable)
1319{
1320 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1321}
1322EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1323
Xiao Guangrongd5661042012-07-17 21:56:16 +08001324pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001325{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001326 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001327}
1328
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001329pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1330{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001331 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001332}
1333EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1334
Xiao Guangrong48987782010-08-22 19:11:43 +08001335int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1336 int nr_pages)
1337{
1338 unsigned long addr;
1339 gfn_t entry;
1340
Gleb Natapov49c77542010-10-18 15:22:23 +02001341 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001342 if (kvm_is_error_hva(addr))
1343 return -1;
1344
1345 if (entry < nr_pages)
1346 return 0;
1347
1348 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1349}
1350EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1351
Xiao Guangronga2766322012-07-26 11:58:59 +08001352static struct page *kvm_pfn_to_page(pfn_t pfn)
1353{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001354 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001355 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001356
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001357 if (kvm_is_mmio_pfn(pfn)) {
1358 WARN_ON(1);
1359 return KVM_ERR_PTR_BAD_PAGE;
1360 }
1361
Xiao Guangronga2766322012-07-26 11:58:59 +08001362 return pfn_to_page(pfn);
1363}
1364
Anthony Liguori35149e22008-04-02 14:46:56 -05001365struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1366{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001367 pfn_t pfn;
1368
1369 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001370
Xiao Guangronga2766322012-07-26 11:58:59 +08001371 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001372}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001373
Avi Kivity954bbbc2007-03-30 14:02:32 +03001374EXPORT_SYMBOL_GPL(gfn_to_page);
1375
Izik Eidusb4231d62007-11-20 11:49:33 +02001376void kvm_release_page_clean(struct page *page)
1377{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001378 WARN_ON(is_error_page(page));
1379
Anthony Liguori35149e22008-04-02 14:46:56 -05001380 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001381}
1382EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1383
Anthony Liguori35149e22008-04-02 14:46:56 -05001384void kvm_release_pfn_clean(pfn_t pfn)
1385{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001386 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001387 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001388}
1389EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1390
Izik Eidusb4231d62007-11-20 11:49:33 +02001391void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001392{
Xiao Guangronga2766322012-07-26 11:58:59 +08001393 WARN_ON(is_error_page(page));
1394
Anthony Liguori35149e22008-04-02 14:46:56 -05001395 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001396}
Izik Eidusb4231d62007-11-20 11:49:33 +02001397EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001398
Anthony Liguori35149e22008-04-02 14:46:56 -05001399void kvm_release_pfn_dirty(pfn_t pfn)
1400{
1401 kvm_set_pfn_dirty(pfn);
1402 kvm_release_pfn_clean(pfn);
1403}
1404EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1405
1406void kvm_set_page_dirty(struct page *page)
1407{
1408 kvm_set_pfn_dirty(page_to_pfn(page));
1409}
1410EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1411
1412void kvm_set_pfn_dirty(pfn_t pfn)
1413{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001414 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001415 struct page *page = pfn_to_page(pfn);
1416 if (!PageReserved(page))
1417 SetPageDirty(page);
1418 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001419}
1420EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1421
1422void kvm_set_pfn_accessed(pfn_t pfn)
1423{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001424 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001425 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001426}
1427EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1428
1429void kvm_get_pfn(pfn_t pfn)
1430{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001431 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001432 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001433}
1434EXPORT_SYMBOL_GPL(kvm_get_pfn);
1435
Izik Eidus195aefd2007-10-01 22:14:18 +02001436static int next_segment(unsigned long len, int offset)
1437{
1438 if (len > PAGE_SIZE - offset)
1439 return PAGE_SIZE - offset;
1440 else
1441 return len;
1442}
1443
1444int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1445 int len)
1446{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001447 int r;
1448 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001449
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001450 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001451 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001452 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001453 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001454 if (r)
1455 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001456 return 0;
1457}
1458EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1459
1460int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1461{
1462 gfn_t gfn = gpa >> PAGE_SHIFT;
1463 int seg;
1464 int offset = offset_in_page(gpa);
1465 int ret;
1466
1467 while ((seg = next_segment(len, offset)) != 0) {
1468 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1469 if (ret < 0)
1470 return ret;
1471 offset = 0;
1472 len -= seg;
1473 data += seg;
1474 ++gfn;
1475 }
1476 return 0;
1477}
1478EXPORT_SYMBOL_GPL(kvm_read_guest);
1479
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001480int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1481 unsigned long len)
1482{
1483 int r;
1484 unsigned long addr;
1485 gfn_t gfn = gpa >> PAGE_SHIFT;
1486 int offset = offset_in_page(gpa);
1487
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001488 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001489 if (kvm_is_error_hva(addr))
1490 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001491 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001492 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001493 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001494 if (r)
1495 return -EFAULT;
1496 return 0;
1497}
1498EXPORT_SYMBOL(kvm_read_guest_atomic);
1499
Izik Eidus195aefd2007-10-01 22:14:18 +02001500int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1501 int offset, int len)
1502{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001503 int r;
1504 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001505
Izik Eiduse0506bc2007-11-11 22:10:22 +02001506 addr = gfn_to_hva(kvm, gfn);
1507 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001508 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001509 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001510 if (r)
1511 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001512 mark_page_dirty(kvm, gfn);
1513 return 0;
1514}
1515EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1516
1517int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1518 unsigned long len)
1519{
1520 gfn_t gfn = gpa >> PAGE_SHIFT;
1521 int seg;
1522 int offset = offset_in_page(gpa);
1523 int ret;
1524
1525 while ((seg = next_segment(len, offset)) != 0) {
1526 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1527 if (ret < 0)
1528 return ret;
1529 offset = 0;
1530 len -= seg;
1531 data += seg;
1532 ++gfn;
1533 }
1534 return 0;
1535}
1536
Gleb Natapov49c77542010-10-18 15:22:23 +02001537int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001538 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001539{
1540 struct kvm_memslots *slots = kvm_memslots(kvm);
1541 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001542 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1543 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1544 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1545 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001546
1547 ghc->gpa = gpa;
1548 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001549 ghc->len = len;
1550 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1551 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1552 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001553 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001554 } else {
1555 /*
1556 * If the requested region crosses two memslots, we still
1557 * verify that the entire region is valid here.
1558 */
1559 while (start_gfn <= end_gfn) {
1560 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1561 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1562 &nr_pages_avail);
1563 if (kvm_is_error_hva(ghc->hva))
1564 return -EFAULT;
1565 start_gfn += nr_pages_avail;
1566 }
1567 /* Use the slow path for cross page reads and writes. */
1568 ghc->memslot = NULL;
1569 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001570 return 0;
1571}
1572EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1573
1574int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1575 void *data, unsigned long len)
1576{
1577 struct kvm_memslots *slots = kvm_memslots(kvm);
1578 int r;
1579
Andrew Honig8f964522013-03-29 09:35:21 -07001580 BUG_ON(len > ghc->len);
1581
Gleb Natapov49c77542010-10-18 15:22:23 +02001582 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001583 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1584
1585 if (unlikely(!ghc->memslot))
1586 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001587
1588 if (kvm_is_error_hva(ghc->hva))
1589 return -EFAULT;
1590
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001591 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001592 if (r)
1593 return -EFAULT;
1594 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1595
1596 return 0;
1597}
1598EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1599
Gleb Natapove03b6442011-07-11 15:28:11 -04001600int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1601 void *data, unsigned long len)
1602{
1603 struct kvm_memslots *slots = kvm_memslots(kvm);
1604 int r;
1605
Andrew Honig8f964522013-03-29 09:35:21 -07001606 BUG_ON(len > ghc->len);
1607
Gleb Natapove03b6442011-07-11 15:28:11 -04001608 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001609 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1610
1611 if (unlikely(!ghc->memslot))
1612 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001613
1614 if (kvm_is_error_hva(ghc->hva))
1615 return -EFAULT;
1616
1617 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1618 if (r)
1619 return -EFAULT;
1620
1621 return 0;
1622}
1623EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1624
Izik Eidus195aefd2007-10-01 22:14:18 +02001625int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1626{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001627 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1628 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001629}
1630EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1631
1632int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1633{
1634 gfn_t gfn = gpa >> PAGE_SHIFT;
1635 int seg;
1636 int offset = offset_in_page(gpa);
1637 int ret;
1638
1639 while ((seg = next_segment(len, offset)) != 0) {
1640 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1641 if (ret < 0)
1642 return ret;
1643 offset = 0;
1644 len -= seg;
1645 ++gfn;
1646 }
1647 return 0;
1648}
1649EXPORT_SYMBOL_GPL(kvm_clear_guest);
1650
Gleb Natapov49c77542010-10-18 15:22:23 +02001651void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1652 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001653{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001654 if (memslot && memslot->dirty_bitmap) {
1655 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001656
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001657 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001658 }
1659}
1660
Gleb Natapov49c77542010-10-18 15:22:23 +02001661void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1662{
1663 struct kvm_memory_slot *memslot;
1664
1665 memslot = gfn_to_memslot(kvm, gfn);
1666 mark_page_dirty_in_slot(kvm, memslot, gfn);
1667}
1668
Eddie Dongb6958ce2007-07-18 12:15:21 +03001669/*
1670 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1671 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001672void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001673{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001674 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001675
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001676 for (;;) {
1677 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001678
Gleb Natapova1b37102009-07-09 15:33:52 +03001679 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001680 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001681 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001682 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001683 if (kvm_cpu_has_pending_timer(vcpu))
1684 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001685 if (signal_pending(current))
1686 break;
1687
Eddie Dongb6958ce2007-07-18 12:15:21 +03001688 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001689 }
1690
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001691 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001692}
1693
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
Avi Kivity6aa8b732006-12-10 02:21:36 -08001719void kvm_resched(struct kvm_vcpu *vcpu)
1720{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001721 if (!need_resched())
1722 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001723 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001724}
1725EXPORT_SYMBOL_GPL(kvm_resched);
1726
Konstantin Weitz41628d32012-04-25 15:30:38 +02001727bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1728{
1729 struct pid *pid;
1730 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301731 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001732
1733 rcu_read_lock();
1734 pid = rcu_dereference(target->pid);
1735 if (pid)
1736 task = get_pid_task(target->pid, PIDTYPE_PID);
1737 rcu_read_unlock();
1738 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301739 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001740 if (task->flags & PF_VCPU) {
1741 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301742 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001743 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301744 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001745 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301746
1747 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001748}
1749EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1750
Raghavendra K T06e48c52012-07-19 15:17:52 +05301751#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1752/*
1753 * Helper that checks whether a VCPU is eligible for directed yield.
1754 * Most eligible candidate to yield is decided by following heuristics:
1755 *
1756 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1757 * (preempted lock holder), indicated by @in_spin_loop.
1758 * Set at the beiginning and cleared at the end of interception/PLE handler.
1759 *
1760 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1761 * chance last time (mostly it has become eligible now since we have probably
1762 * yielded to lockholder in last iteration. This is done by toggling
1763 * @dy_eligible each time a VCPU checked for eligibility.)
1764 *
1765 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1766 * to preempted lock-holder could result in wrong VCPU selection and CPU
1767 * burning. Giving priority for a potential lock-holder increases lock
1768 * progress.
1769 *
1770 * Since algorithm is based on heuristics, accessing another VCPU data without
1771 * locking does not harm. It may result in trying to yield to same VCPU, fail
1772 * and continue with next VCPU and so on.
1773 */
1774bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1775{
1776 bool eligible;
1777
1778 eligible = !vcpu->spin_loop.in_spin_loop ||
1779 (vcpu->spin_loop.in_spin_loop &&
1780 vcpu->spin_loop.dy_eligible);
1781
1782 if (vcpu->spin_loop.in_spin_loop)
1783 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1784
1785 return eligible;
1786}
1787#endif
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301788
Rik van Riel217ece62011-02-01 09:53:28 -05001789void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001790{
Rik van Riel217ece62011-02-01 09:53:28 -05001791 struct kvm *kvm = me->kvm;
1792 struct kvm_vcpu *vcpu;
1793 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1794 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301795 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001796 int pass;
1797 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001798
Raghavendra K T4c088492012-07-18 19:07:46 +05301799 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001800 /*
1801 * We boost the priority of a VCPU that is runnable but not
1802 * currently running, because it got preempted by something
1803 * else and called schedule in __vcpu_run. Hopefully that
1804 * VCPU is holding the lock that we need and will release it.
1805 * We approximate round-robin by starting at the last boosted VCPU.
1806 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301807 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001808 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001809 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001810 i = last_boosted_vcpu;
1811 continue;
1812 } else if (pass && i > last_boosted_vcpu)
1813 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301814 if (!ACCESS_ONCE(vcpu->preempted))
1815 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001816 if (vcpu == me)
1817 continue;
1818 if (waitqueue_active(&vcpu->wq))
1819 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301820 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1821 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301822
1823 yielded = kvm_vcpu_yield_to(vcpu);
1824 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001825 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001826 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301827 } else if (yielded < 0) {
1828 try--;
1829 if (!try)
1830 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001831 }
Rik van Riel217ece62011-02-01 09:53:28 -05001832 }
1833 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301834 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301835
1836 /* Ensure vcpu is not eligible during next spinloop */
1837 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001838}
1839EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1840
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001841static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001842{
1843 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001844 struct page *page;
1845
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001846 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001847 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001848#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001849 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001850 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001851#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001852#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1853 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1854 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1855#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001856 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001857 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001858 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001859 vmf->page = page;
1860 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001861}
1862
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001863static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001864 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001865};
1866
1867static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1868{
1869 vma->vm_ops = &kvm_vcpu_vm_ops;
1870 return 0;
1871}
1872
Avi Kivitybccf2152007-02-21 18:04:26 +02001873static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1874{
1875 struct kvm_vcpu *vcpu = filp->private_data;
1876
Al Viro66c0b392008-04-19 20:33:56 +01001877 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001878 return 0;
1879}
1880
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001881static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001882 .release = kvm_vcpu_release,
1883 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001884#ifdef CONFIG_COMPAT
1885 .compat_ioctl = kvm_vcpu_compat_ioctl,
1886#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001887 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001888 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001889};
1890
1891/*
1892 * Allocates an inode for the vcpu.
1893 */
1894static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1895{
Roland Dreier628ff7c2009-12-18 09:41:24 -08001896 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
Avi Kivitybccf2152007-02-21 18:04:26 +02001897}
1898
Avi Kivityc5ea7662007-02-20 18:41:05 +02001899/*
1900 * Creates some virtual cpus. Good luck creating more than one.
1901 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001902static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001903{
1904 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001905 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001906
Gleb Natapov73880c82009-06-09 15:56:28 +03001907 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001908 if (IS_ERR(vcpu))
1909 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001910
Avi Kivity15ad7142007-07-11 18:17:21 +03001911 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1912
Avi Kivity26e52152007-11-20 15:30:24 +02001913 r = kvm_arch_vcpu_setup(vcpu);
1914 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001915 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001916
Shaohua Li11ec2802007-07-23 14:51:37 +08001917 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001918 if (!kvm_vcpu_compatible(vcpu)) {
1919 r = -EINVAL;
1920 goto unlock_vcpu_destroy;
1921 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001922 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1923 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001924 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001925 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001926
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001927 kvm_for_each_vcpu(r, v, kvm)
1928 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001929 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001930 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001931 }
1932
1933 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001934
1935 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001936 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001937 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001938 if (r < 0) {
1939 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001940 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001941 }
1942
1943 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1944 smp_wmb();
1945 atomic_inc(&kvm->online_vcpus);
1946
Gleb Natapov73880c82009-06-09 15:56:28 +03001947 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001948 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001949 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001950
Jan Kiszkad7805922011-05-23 10:33:05 +02001951unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001952 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001953vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001954 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001955 return r;
1956}
1957
Avi Kivity1961d272007-03-05 19:46:05 +02001958static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1959{
1960 if (sigset) {
1961 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1962 vcpu->sigset_active = 1;
1963 vcpu->sigset = *sigset;
1964 } else
1965 vcpu->sigset_active = 0;
1966 return 0;
1967}
1968
Avi Kivitybccf2152007-02-21 18:04:26 +02001969static long kvm_vcpu_ioctl(struct file *filp,
1970 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001971{
Avi Kivitybccf2152007-02-21 18:04:26 +02001972 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001973 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001974 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001975 struct kvm_fpu *fpu = NULL;
1976 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001977
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001978 if (vcpu->kvm->mm != current->mm)
1979 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001980
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08001981#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03001982 /*
1983 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1984 * so vcpu_load() would break it.
1985 */
1986 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1987 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1988#endif
1989
1990
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001991 r = vcpu_load(vcpu);
1992 if (r)
1993 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001995 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001996 r = -EINVAL;
1997 if (arg)
1998 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001999 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002000 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002001 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002002 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002003 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002004
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002005 r = -ENOMEM;
2006 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2007 if (!kvm_regs)
2008 goto out;
2009 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002010 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002011 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002012 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002013 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2014 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002015 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002016out_free1:
2017 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002018 break;
2019 }
2020 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002021 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002022
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002023 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002024 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2025 if (IS_ERR(kvm_regs)) {
2026 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002027 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002028 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002029 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002030 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002031 break;
2032 }
2033 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002034 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2035 r = -ENOMEM;
2036 if (!kvm_sregs)
2037 goto out;
2038 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002039 if (r)
2040 goto out;
2041 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002042 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002043 goto out;
2044 r = 0;
2045 break;
2046 }
2047 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002048 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2049 if (IS_ERR(kvm_sregs)) {
2050 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002051 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002052 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002053 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002054 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002055 break;
2056 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002057 case KVM_GET_MP_STATE: {
2058 struct kvm_mp_state mp_state;
2059
2060 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2061 if (r)
2062 goto out;
2063 r = -EFAULT;
2064 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2065 goto out;
2066 r = 0;
2067 break;
2068 }
2069 case KVM_SET_MP_STATE: {
2070 struct kvm_mp_state mp_state;
2071
2072 r = -EFAULT;
2073 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2074 goto out;
2075 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002076 break;
2077 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002078 case KVM_TRANSLATE: {
2079 struct kvm_translation tr;
2080
2081 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002082 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002083 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002084 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002085 if (r)
2086 goto out;
2087 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002088 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002089 goto out;
2090 r = 0;
2091 break;
2092 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002093 case KVM_SET_GUEST_DEBUG: {
2094 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002095
2096 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002097 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002098 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002099 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002100 break;
2101 }
Avi Kivity1961d272007-03-05 19:46:05 +02002102 case KVM_SET_SIGNAL_MASK: {
2103 struct kvm_signal_mask __user *sigmask_arg = argp;
2104 struct kvm_signal_mask kvm_sigmask;
2105 sigset_t sigset, *p;
2106
2107 p = NULL;
2108 if (argp) {
2109 r = -EFAULT;
2110 if (copy_from_user(&kvm_sigmask, argp,
2111 sizeof kvm_sigmask))
2112 goto out;
2113 r = -EINVAL;
2114 if (kvm_sigmask.len != sizeof sigset)
2115 goto out;
2116 r = -EFAULT;
2117 if (copy_from_user(&sigset, sigmask_arg->sigset,
2118 sizeof sigset))
2119 goto out;
2120 p = &sigset;
2121 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002122 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002123 break;
2124 }
Avi Kivityb8836732007-04-01 16:34:31 +03002125 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002126 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2127 r = -ENOMEM;
2128 if (!fpu)
2129 goto out;
2130 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002131 if (r)
2132 goto out;
2133 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002134 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002135 goto out;
2136 r = 0;
2137 break;
2138 }
2139 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002140 fpu = memdup_user(argp, sizeof(*fpu));
2141 if (IS_ERR(fpu)) {
2142 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002143 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002144 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002145 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002146 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002147 break;
2148 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002149 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002150 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002151 }
2152out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002153 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002154 kfree(fpu);
2155 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002156 return r;
2157}
2158
Alexander Graf1dda6062011-06-08 02:45:37 +02002159#ifdef CONFIG_COMPAT
2160static long kvm_vcpu_compat_ioctl(struct file *filp,
2161 unsigned int ioctl, unsigned long arg)
2162{
2163 struct kvm_vcpu *vcpu = filp->private_data;
2164 void __user *argp = compat_ptr(arg);
2165 int r;
2166
2167 if (vcpu->kvm->mm != current->mm)
2168 return -EIO;
2169
2170 switch (ioctl) {
2171 case KVM_SET_SIGNAL_MASK: {
2172 struct kvm_signal_mask __user *sigmask_arg = argp;
2173 struct kvm_signal_mask kvm_sigmask;
2174 compat_sigset_t csigset;
2175 sigset_t sigset;
2176
2177 if (argp) {
2178 r = -EFAULT;
2179 if (copy_from_user(&kvm_sigmask, argp,
2180 sizeof kvm_sigmask))
2181 goto out;
2182 r = -EINVAL;
2183 if (kvm_sigmask.len != sizeof csigset)
2184 goto out;
2185 r = -EFAULT;
2186 if (copy_from_user(&csigset, sigmask_arg->sigset,
2187 sizeof csigset))
2188 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002189 sigset_from_compat(&sigset, &csigset);
2190 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2191 } else
2192 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002193 break;
2194 }
2195 default:
2196 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2197 }
2198
2199out:
2200 return r;
2201}
2202#endif
2203
Scott Wood852b6d52013-04-12 14:08:42 +00002204static int kvm_device_ioctl_attr(struct kvm_device *dev,
2205 int (*accessor)(struct kvm_device *dev,
2206 struct kvm_device_attr *attr),
2207 unsigned long arg)
2208{
2209 struct kvm_device_attr attr;
2210
2211 if (!accessor)
2212 return -EPERM;
2213
2214 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2215 return -EFAULT;
2216
2217 return accessor(dev, &attr);
2218}
2219
2220static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2221 unsigned long arg)
2222{
2223 struct kvm_device *dev = filp->private_data;
2224
2225 switch (ioctl) {
2226 case KVM_SET_DEVICE_ATTR:
2227 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2228 case KVM_GET_DEVICE_ATTR:
2229 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2230 case KVM_HAS_DEVICE_ATTR:
2231 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2232 default:
2233 if (dev->ops->ioctl)
2234 return dev->ops->ioctl(dev, ioctl, arg);
2235
2236 return -ENOTTY;
2237 }
2238}
2239
Scott Wood852b6d52013-04-12 14:08:42 +00002240static int kvm_device_release(struct inode *inode, struct file *filp)
2241{
2242 struct kvm_device *dev = filp->private_data;
2243 struct kvm *kvm = dev->kvm;
2244
Scott Wood852b6d52013-04-12 14:08:42 +00002245 kvm_put_kvm(kvm);
2246 return 0;
2247}
2248
2249static const struct file_operations kvm_device_fops = {
2250 .unlocked_ioctl = kvm_device_ioctl,
Scott Wooddb6ae612013-04-30 20:00:45 -05002251#ifdef CONFIG_COMPAT
2252 .compat_ioctl = kvm_device_ioctl,
2253#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002254 .release = kvm_device_release,
2255};
2256
2257struct kvm_device *kvm_device_from_filp(struct file *filp)
2258{
2259 if (filp->f_op != &kvm_device_fops)
2260 return NULL;
2261
2262 return filp->private_data;
2263}
2264
2265static int kvm_ioctl_create_device(struct kvm *kvm,
2266 struct kvm_create_device *cd)
2267{
2268 struct kvm_device_ops *ops = NULL;
2269 struct kvm_device *dev;
2270 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2271 int ret;
2272
2273 switch (cd->type) {
Scott Wood5df554a2013-04-12 14:08:46 +00002274#ifdef CONFIG_KVM_MPIC
2275 case KVM_DEV_TYPE_FSL_MPIC_20:
2276 case KVM_DEV_TYPE_FSL_MPIC_42:
2277 ops = &kvm_mpic_ops;
2278 break;
2279#endif
Paul Mackerras5975a2e2013-04-27 00:28:37 +00002280#ifdef CONFIG_KVM_XICS
2281 case KVM_DEV_TYPE_XICS:
2282 ops = &kvm_xics_ops;
2283 break;
2284#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002285 default:
2286 return -ENODEV;
2287 }
2288
2289 if (test)
2290 return 0;
2291
2292 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2293 if (!dev)
2294 return -ENOMEM;
2295
2296 dev->ops = ops;
2297 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002298
2299 ret = ops->create(dev, cd->type);
2300 if (ret < 0) {
2301 kfree(dev);
2302 return ret;
2303 }
2304
2305 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR);
2306 if (ret < 0) {
2307 ops->destroy(dev);
2308 return ret;
2309 }
2310
Scott Wood07f0a7b2013-04-25 14:11:23 +00002311 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002312 kvm_get_kvm(kvm);
2313 cd->fd = ret;
2314 return 0;
2315}
2316
Avi Kivitybccf2152007-02-21 18:04:26 +02002317static long kvm_vm_ioctl(struct file *filp,
2318 unsigned int ioctl, unsigned long arg)
2319{
2320 struct kvm *kvm = filp->private_data;
2321 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002322 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002323
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002324 if (kvm->mm != current->mm)
2325 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002326 switch (ioctl) {
2327 case KVM_CREATE_VCPU:
2328 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002329 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002330 case KVM_SET_USER_MEMORY_REGION: {
2331 struct kvm_userspace_memory_region kvm_userspace_mem;
2332
2333 r = -EFAULT;
2334 if (copy_from_user(&kvm_userspace_mem, argp,
2335 sizeof kvm_userspace_mem))
2336 goto out;
2337
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002338 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002339 break;
2340 }
2341 case KVM_GET_DIRTY_LOG: {
2342 struct kvm_dirty_log log;
2343
2344 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002345 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002346 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002347 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002348 break;
2349 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002350#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2351 case KVM_REGISTER_COALESCED_MMIO: {
2352 struct kvm_coalesced_mmio_zone zone;
2353 r = -EFAULT;
2354 if (copy_from_user(&zone, argp, sizeof zone))
2355 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002356 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002357 break;
2358 }
2359 case KVM_UNREGISTER_COALESCED_MMIO: {
2360 struct kvm_coalesced_mmio_zone zone;
2361 r = -EFAULT;
2362 if (copy_from_user(&zone, argp, sizeof zone))
2363 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002364 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002365 break;
2366 }
2367#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002368 case KVM_IRQFD: {
2369 struct kvm_irqfd data;
2370
2371 r = -EFAULT;
2372 if (copy_from_user(&data, argp, sizeof data))
2373 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002374 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002375 break;
2376 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002377 case KVM_IOEVENTFD: {
2378 struct kvm_ioeventfd data;
2379
2380 r = -EFAULT;
2381 if (copy_from_user(&data, argp, sizeof data))
2382 goto out;
2383 r = kvm_ioeventfd(kvm, &data);
2384 break;
2385 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002386#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2387 case KVM_SET_BOOT_CPU_ID:
2388 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002389 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002390 if (atomic_read(&kvm->online_vcpus) != 0)
2391 r = -EBUSY;
2392 else
2393 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002394 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002395 break;
2396#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002397#ifdef CONFIG_HAVE_KVM_MSI
2398 case KVM_SIGNAL_MSI: {
2399 struct kvm_msi msi;
2400
2401 r = -EFAULT;
2402 if (copy_from_user(&msi, argp, sizeof msi))
2403 goto out;
2404 r = kvm_send_userspace_msi(kvm, &msi);
2405 break;
2406 }
2407#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002408#ifdef __KVM_HAVE_IRQ_LINE
2409 case KVM_IRQ_LINE_STATUS:
2410 case KVM_IRQ_LINE: {
2411 struct kvm_irq_level irq_event;
2412
2413 r = -EFAULT;
2414 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2415 goto out;
2416
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002417 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2418 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002419 if (r)
2420 goto out;
2421
2422 r = -EFAULT;
2423 if (ioctl == KVM_IRQ_LINE_STATUS) {
2424 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2425 goto out;
2426 }
2427
2428 r = 0;
2429 break;
2430 }
2431#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002432#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2433 case KVM_SET_GSI_ROUTING: {
2434 struct kvm_irq_routing routing;
2435 struct kvm_irq_routing __user *urouting;
2436 struct kvm_irq_routing_entry *entries;
2437
2438 r = -EFAULT;
2439 if (copy_from_user(&routing, argp, sizeof(routing)))
2440 goto out;
2441 r = -EINVAL;
2442 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2443 goto out;
2444 if (routing.flags)
2445 goto out;
2446 r = -ENOMEM;
2447 entries = vmalloc(routing.nr * sizeof(*entries));
2448 if (!entries)
2449 goto out;
2450 r = -EFAULT;
2451 urouting = argp;
2452 if (copy_from_user(entries, urouting->entries,
2453 routing.nr * sizeof(*entries)))
2454 goto out_free_irq_routing;
2455 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2456 routing.flags);
2457 out_free_irq_routing:
2458 vfree(entries);
2459 break;
2460 }
2461#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002462 case KVM_CREATE_DEVICE: {
2463 struct kvm_create_device cd;
2464
2465 r = -EFAULT;
2466 if (copy_from_user(&cd, argp, sizeof(cd)))
2467 goto out;
2468
2469 r = kvm_ioctl_create_device(kvm, &cd);
2470 if (r)
2471 goto out;
2472
2473 r = -EFAULT;
2474 if (copy_to_user(argp, &cd, sizeof(cd)))
2475 goto out;
2476
2477 r = 0;
2478 break;
2479 }
Avi Kivityf17abe92007-02-21 19:28:04 +02002480 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002481 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002482 if (r == -ENOTTY)
2483 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002484 }
2485out:
2486 return r;
2487}
2488
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002489#ifdef CONFIG_COMPAT
2490struct compat_kvm_dirty_log {
2491 __u32 slot;
2492 __u32 padding1;
2493 union {
2494 compat_uptr_t dirty_bitmap; /* one bit per page */
2495 __u64 padding2;
2496 };
2497};
2498
2499static long kvm_vm_compat_ioctl(struct file *filp,
2500 unsigned int ioctl, unsigned long arg)
2501{
2502 struct kvm *kvm = filp->private_data;
2503 int r;
2504
2505 if (kvm->mm != current->mm)
2506 return -EIO;
2507 switch (ioctl) {
2508 case KVM_GET_DIRTY_LOG: {
2509 struct compat_kvm_dirty_log compat_log;
2510 struct kvm_dirty_log log;
2511
2512 r = -EFAULT;
2513 if (copy_from_user(&compat_log, (void __user *)arg,
2514 sizeof(compat_log)))
2515 goto out;
2516 log.slot = compat_log.slot;
2517 log.padding1 = compat_log.padding1;
2518 log.padding2 = compat_log.padding2;
2519 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2520
2521 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002522 break;
2523 }
2524 default:
2525 r = kvm_vm_ioctl(filp, ioctl, arg);
2526 }
2527
2528out:
2529 return r;
2530}
2531#endif
2532
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002533static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002534{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002535 struct page *page[1];
2536 unsigned long addr;
2537 int npages;
2538 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002539 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002540
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002541 addr = gfn_to_hva(kvm, gfn);
2542 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002543 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002544
2545 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2546 NULL);
2547 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002548 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002549
2550 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002551 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002552}
2553
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002554static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002555 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002556};
2557
2558static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2559{
2560 vma->vm_ops = &kvm_vm_vm_ops;
2561 return 0;
2562}
2563
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002564static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002565 .release = kvm_vm_release,
2566 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002567#ifdef CONFIG_COMPAT
2568 .compat_ioctl = kvm_vm_compat_ioctl,
2569#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002570 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002571 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002572};
2573
Carsten Ottee08b9632012-01-04 10:25:20 +01002574static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002575{
Heiko Carstensaac87632010-10-27 17:22:10 +02002576 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002577 struct kvm *kvm;
2578
Carsten Ottee08b9632012-01-04 10:25:20 +01002579 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002580 if (IS_ERR(kvm))
2581 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002582#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2583 r = kvm_coalesced_mmio_init(kvm);
2584 if (r < 0) {
2585 kvm_put_kvm(kvm);
2586 return r;
2587 }
2588#endif
Heiko Carstensaac87632010-10-27 17:22:10 +02002589 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2590 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002591 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002592
Heiko Carstensaac87632010-10-27 17:22:10 +02002593 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002594}
2595
Avi Kivity1a811b62008-12-08 18:25:27 +02002596static long kvm_dev_ioctl_check_extension_generic(long arg)
2597{
2598 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002599 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002600 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002601 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002602#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2603 case KVM_CAP_SET_BOOT_CPU_ID:
2604#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002605 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002606#ifdef CONFIG_HAVE_KVM_MSI
2607 case KVM_CAP_SIGNAL_MSI:
2608#endif
Alexander Graf7df35f52013-04-16 12:12:49 +02002609#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2610 case KVM_CAP_IRQFD_RESAMPLE:
2611#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002612 return 1;
Alexander Grafa725d562013-04-17 13:29:30 +02002613#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002614 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002615 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002616#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002617 default:
2618 break;
2619 }
2620 return kvm_dev_ioctl_check_extension(arg);
2621}
2622
Avi Kivityf17abe92007-02-21 19:28:04 +02002623static long kvm_dev_ioctl(struct file *filp,
2624 unsigned int ioctl, unsigned long arg)
2625{
Avi Kivity07c45a32007-03-07 13:05:38 +02002626 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002627
2628 switch (ioctl) {
2629 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002630 r = -EINVAL;
2631 if (arg)
2632 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002633 r = KVM_API_VERSION;
2634 break;
2635 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002636 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002637 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002638 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002639 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002640 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002641 case KVM_GET_VCPU_MMAP_SIZE:
2642 r = -EINVAL;
2643 if (arg)
2644 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002645 r = PAGE_SIZE; /* struct kvm_run */
2646#ifdef CONFIG_X86
2647 r += PAGE_SIZE; /* pio data page */
2648#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002649#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2650 r += PAGE_SIZE; /* coalesced mmio ring page */
2651#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002652 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002653 case KVM_TRACE_ENABLE:
2654 case KVM_TRACE_PAUSE:
2655 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002656 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002657 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002658 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002659 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002660 }
2661out:
2662 return r;
2663}
2664
Avi Kivity6aa8b732006-12-10 02:21:36 -08002665static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002666 .unlocked_ioctl = kvm_dev_ioctl,
2667 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002668 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002669};
2670
2671static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002672 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002673 "kvm",
2674 &kvm_chardev_ops,
2675};
2676
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002677static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002678{
2679 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002680 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002681
Rusty Russell7f59f492008-12-07 21:25:45 +10302682 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002683 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002684
Rusty Russell7f59f492008-12-07 21:25:45 +10302685 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002686
2687 r = kvm_arch_hardware_enable(NULL);
2688
2689 if (r) {
2690 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2691 atomic_inc(&hardware_enable_failed);
2692 printk(KERN_INFO "kvm: enabling virtualization on "
2693 "CPU%d failed\n", cpu);
2694 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002695}
2696
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002697static void hardware_enable(void *junk)
2698{
Jan Kiszkae935b832011-02-08 12:55:33 +01002699 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002700 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002701 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002702}
2703
2704static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002705{
2706 int cpu = raw_smp_processor_id();
2707
Rusty Russell7f59f492008-12-07 21:25:45 +10302708 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002709 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302710 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002711 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002712}
2713
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002714static void hardware_disable(void *junk)
2715{
Jan Kiszkae935b832011-02-08 12:55:33 +01002716 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002717 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002718 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002719}
2720
Alexander Graf10474ae2009-09-15 11:37:46 +02002721static void hardware_disable_all_nolock(void)
2722{
2723 BUG_ON(!kvm_usage_count);
2724
2725 kvm_usage_count--;
2726 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002727 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002728}
2729
2730static void hardware_disable_all(void)
2731{
Jan Kiszkae935b832011-02-08 12:55:33 +01002732 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002733 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002734 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002735}
2736
2737static int hardware_enable_all(void)
2738{
2739 int r = 0;
2740
Jan Kiszkae935b832011-02-08 12:55:33 +01002741 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002742
2743 kvm_usage_count++;
2744 if (kvm_usage_count == 1) {
2745 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002746 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002747
2748 if (atomic_read(&hardware_enable_failed)) {
2749 hardware_disable_all_nolock();
2750 r = -EBUSY;
2751 }
2752 }
2753
Jan Kiszkae935b832011-02-08 12:55:33 +01002754 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002755
2756 return r;
2757}
2758
Avi Kivity774c47f2007-02-12 00:54:47 -08002759static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2760 void *v)
2761{
2762 int cpu = (long)v;
2763
Alexander Graf10474ae2009-09-15 11:37:46 +02002764 if (!kvm_usage_count)
2765 return NOTIFY_OK;
2766
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002767 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002768 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002769 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002770 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2771 cpu);
2772 hardware_disable(NULL);
2773 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002774 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002775 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2776 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002777 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002778 break;
2779 }
2780 return NOTIFY_OK;
2781}
2782
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002783static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002784 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002785{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002786 /*
2787 * Some (well, at least mine) BIOSes hang on reboot if
2788 * in vmx root mode.
2789 *
2790 * And Intel TXT required VMX off for all cpu when system shutdown.
2791 */
2792 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2793 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002794 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002795 return NOTIFY_OK;
2796}
2797
2798static struct notifier_block kvm_reboot_notifier = {
2799 .notifier_call = kvm_reboot,
2800 .priority = 0,
2801};
2802
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002803static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002804{
2805 int i;
2806
2807 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002808 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002809
2810 kvm_iodevice_destructor(pos);
2811 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002812 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002813}
2814
Geoff Levand39369f72013-04-05 19:20:30 +00002815static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002816{
2817 const struct kvm_io_range *r1 = p1;
2818 const struct kvm_io_range *r2 = p2;
2819
2820 if (r1->addr < r2->addr)
2821 return -1;
2822 if (r1->addr + r1->len > r2->addr + r2->len)
2823 return 1;
2824 return 0;
2825}
2826
Geoff Levand39369f72013-04-05 19:20:30 +00002827static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002828 gpa_t addr, int len)
2829{
Sasha Levin743eeb02011-07-27 16:00:48 +03002830 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2831 .addr = addr,
2832 .len = len,
2833 .dev = dev,
2834 };
2835
2836 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2837 kvm_io_bus_sort_cmp, NULL);
2838
2839 return 0;
2840}
2841
Geoff Levand39369f72013-04-05 19:20:30 +00002842static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002843 gpa_t addr, int len)
2844{
2845 struct kvm_io_range *range, key;
2846 int off;
2847
2848 key = (struct kvm_io_range) {
2849 .addr = addr,
2850 .len = len,
2851 };
2852
2853 range = bsearch(&key, bus->range, bus->dev_count,
2854 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2855 if (range == NULL)
2856 return -ENOENT;
2857
2858 off = range - bus->range;
2859
2860 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2861 off--;
2862
2863 return off;
2864}
2865
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002866/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002867int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002868 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002869{
Sasha Levin743eeb02011-07-27 16:00:48 +03002870 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002871 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002872 struct kvm_io_range range;
2873
2874 range = (struct kvm_io_range) {
2875 .addr = addr,
2876 .len = len,
2877 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002878
2879 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002880 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2881 if (idx < 0)
2882 return -EOPNOTSUPP;
2883
2884 while (idx < bus->dev_count &&
2885 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2886 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002887 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002888 idx++;
2889 }
2890
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002891 return -EOPNOTSUPP;
2892}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002893
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002894/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002895int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2896 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002897{
Sasha Levin743eeb02011-07-27 16:00:48 +03002898 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002899 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002900 struct kvm_io_range range;
2901
2902 range = (struct kvm_io_range) {
2903 .addr = addr,
2904 .len = len,
2905 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002906
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002907 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002908 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2909 if (idx < 0)
2910 return -EOPNOTSUPP;
2911
2912 while (idx < bus->dev_count &&
2913 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2914 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002915 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002916 idx++;
2917 }
2918
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002919 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002920}
2921
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002922/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002923int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2924 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002925{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002926 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002927
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002928 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08002929 /* exclude ioeventfd which is limited by maximum fd */
2930 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002931 return -ENOSPC;
2932
Amos Konga13007162012-03-09 12:17:32 +08002933 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2934 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002935 if (!new_bus)
2936 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002937 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2938 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002939 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002940 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2941 synchronize_srcu_expedited(&kvm->srcu);
2942 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002943
2944 return 0;
2945}
2946
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002947/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002948int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2949 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002950{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002951 int i, r;
2952 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002953
Sasha Levincdfca7b2011-12-04 19:36:28 +02002954 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002955 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002956 for (i = 0; i < bus->dev_count; i++)
2957 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002958 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002959 break;
2960 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002961
Amos Konga13007162012-03-09 12:17:32 +08002962 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002963 return r;
Amos Konga13007162012-03-09 12:17:32 +08002964
2965 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2966 sizeof(struct kvm_io_range)), GFP_KERNEL);
2967 if (!new_bus)
2968 return -ENOMEM;
2969
2970 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2971 new_bus->dev_count--;
2972 memcpy(new_bus->range + i, bus->range + i + 1,
2973 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002974
2975 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2976 synchronize_srcu_expedited(&kvm->srcu);
2977 kfree(bus);
2978 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002979}
2980
Avi Kivity774c47f2007-02-12 00:54:47 -08002981static struct notifier_block kvm_cpu_notifier = {
2982 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08002983};
2984
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002985static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002986{
2987 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002988 struct kvm *kvm;
2989
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002990 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002991 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02002992 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002993 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01002994 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002995 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002996}
2997
2998DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2999
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003000static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003001{
3002 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003003 struct kvm *kvm;
3004 struct kvm_vcpu *vcpu;
3005 int i;
3006
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003007 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01003008 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003009 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003010 kvm_for_each_vcpu(i, vcpu, kvm)
3011 *val += *(u32 *)((void *)vcpu + offset);
3012
Jan Kiszkae935b832011-02-08 12:55:33 +01003013 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003014 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003015}
3016
Avi Kivityba1389b2007-11-18 16:24:12 +02003017DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3018
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003019static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003020 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3021 [KVM_STAT_VM] = &vm_stat_fops,
3022};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003023
Hamo4f69b682011-12-15 14:23:16 +08003024static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003025{
Hamo4f69b682011-12-15 14:23:16 +08003026 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003027 struct kvm_stats_debugfs_item *p;
3028
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003029 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003030 if (kvm_debugfs_dir == NULL)
3031 goto out;
3032
3033 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003034 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003035 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003036 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003037 if (p->dentry == NULL)
3038 goto out_dir;
3039 }
3040
3041 return 0;
3042
3043out_dir:
3044 debugfs_remove_recursive(kvm_debugfs_dir);
3045out:
3046 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003047}
3048
3049static void kvm_exit_debug(void)
3050{
3051 struct kvm_stats_debugfs_item *p;
3052
3053 for (p = debugfs_entries; p->name; ++p)
3054 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003055 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003056}
3057
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003058static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003059{
Alexander Graf10474ae2009-09-15 11:37:46 +02003060 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003061 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003062 return 0;
3063}
3064
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003065static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003066{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003067 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01003068 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003069 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003070 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003071}
3072
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003073static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003074 .suspend = kvm_suspend,
3075 .resume = kvm_resume,
3076};
3077
Avi Kivity15ad7142007-07-11 18:17:21 +03003078static inline
3079struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3080{
3081 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3082}
3083
3084static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3085{
3086 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303087 if (vcpu->preempted)
3088 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003089
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003090 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003091}
3092
3093static void kvm_sched_out(struct preempt_notifier *pn,
3094 struct task_struct *next)
3095{
3096 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3097
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303098 if (current->state == TASK_RUNNING)
3099 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003100 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003101}
3102
Avi Kivity0ee75be2010-04-28 15:39:01 +03003103int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003104 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003105{
3106 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003107 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003108
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003109 r = kvm_arch_init(opaque);
3110 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003111 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003112
Asias He7dac16c2013-05-08 10:57:29 +08003113 /*
3114 * kvm_arch_init makes sure there's at most one caller
3115 * for architectures that support multiple implementations,
3116 * like intel and amd on x86.
3117 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3118 * conflicts in case kvm is already setup for another implementation.
3119 */
3120 r = kvm_irqfd_init();
3121 if (r)
3122 goto out_irqfd;
3123
Avi Kivity8437a612009-06-06 14:52:35 -07003124 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303125 r = -ENOMEM;
3126 goto out_free_0;
3127 }
3128
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003129 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003130 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303131 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003132
Yang, Sheng002c7f72007-07-31 14:23:01 +03003133 for_each_online_cpu(cpu) {
3134 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003135 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003136 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003137 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003138 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003139 }
3140
Avi Kivity774c47f2007-02-12 00:54:47 -08003141 r = register_cpu_notifier(&kvm_cpu_notifier);
3142 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003143 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003144 register_reboot_notifier(&kvm_reboot_notifier);
3145
Rusty Russellc16f8622007-07-30 21:12:19 +10003146 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003147 if (!vcpu_align)
3148 vcpu_align = __alignof__(struct kvm_vcpu);
3149 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003150 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003151 if (!kvm_vcpu_cache) {
3152 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003153 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003154 }
3155
Gleb Natapovaf585b92010-10-14 11:22:46 +02003156 r = kvm_async_pf_init();
3157 if (r)
3158 goto out_free;
3159
Avi Kivity6aa8b732006-12-10 02:21:36 -08003160 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003161 kvm_vm_fops.owner = module;
3162 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003163
3164 r = misc_register(&kvm_dev);
3165 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003166 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003167 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003168 }
3169
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003170 register_syscore_ops(&kvm_syscore_ops);
3171
Avi Kivity15ad7142007-07-11 18:17:21 +03003172 kvm_preempt_ops.sched_in = kvm_sched_in;
3173 kvm_preempt_ops.sched_out = kvm_sched_out;
3174
Hamo4f69b682011-12-15 14:23:16 +08003175 r = kvm_init_debug();
3176 if (r) {
3177 printk(KERN_ERR "kvm: create debugfs files failed\n");
3178 goto out_undebugfs;
3179 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003180
Avi Kivityc7addb92007-09-16 18:58:32 +02003181 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003182
Hamo4f69b682011-12-15 14:23:16 +08003183out_undebugfs:
3184 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003185 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003186out_unreg:
3187 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003188out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003189 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003190out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003191 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003192 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003193out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003194out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003195 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303196out_free_0a:
3197 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003198out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003199 kvm_irqfd_exit();
3200out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003201 kvm_arch_exit();
3202out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003203 return r;
3204}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003205EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003206
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003207void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003208{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003209 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003210 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003211 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003212 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003213 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003214 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003215 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003216 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003217 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003218 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003219 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303220 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003221}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003222EXPORT_SYMBOL_GPL(kvm_exit);