blob: cc6a25d95fbff532bf5b00b0c339bec91ddc5bcf [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
David Matlack2ea75be2014-09-19 16:03:25 -070055#include <asm/ioctl.h>
Avi Kivitye4956062007-06-28 14:15:57 -040056#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020057#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080058
Laurent Vivier5f94c172008-05-30 16:05:54 +020059#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020060#include "async_pf.h"
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +020061#include "vfio.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020062
Marcelo Tosatti229456f2009-06-17 09:22:14 -030063#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h>
65
Avi Kivity6aa8b732006-12-10 02:21:36 -080066MODULE_AUTHOR("Qumranet");
67MODULE_LICENSE("GPL");
68
Paolo Bonzinif7819512015-02-04 18:20:58 +010069unsigned int halt_poll_ns = 0;
70module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
71
Marcelo Tosattifa40a822009-06-04 15:08:24 -030072/*
73 * Ordering of locks:
74 *
Sheng Yangfae3a352009-12-15 10:28:07 +080075 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030076 */
77
Paolo Bonzini2f303b72013-09-25 13:53:07 +020078DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020079static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080080LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080081
Rusty Russell7f59f492008-12-07 21:25:45 +103082static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020083static int kvm_usage_count = 0;
84static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030085
Rusty Russellc16f8622007-07-30 21:12:19 +100086struct kmem_cache *kvm_vcpu_cache;
87EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030088
Avi Kivity15ad7142007-07-11 18:17:21 +030089static __read_mostly struct preempt_ops kvm_preempt_ops;
90
Hollis Blanchard76f7c872008-04-15 16:05:42 -050091struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080092
Avi Kivitybccf2152007-02-21 18:04:26 +020093static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
94 unsigned long arg);
Christian Borntraegerde8e5d72015-02-03 09:35:15 +010095#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +020096static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
97 unsigned long arg);
98#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020099static int hardware_enable_all(void);
100static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +0200101
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200102static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -0800103
104static void kvm_release_pfn_dirty(pfn_t pfn);
105static void mark_page_dirty_in_slot(struct kvm *kvm,
106 struct kvm_memory_slot *memslot, gfn_t gfn);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200107
Andi Kleen52480132014-02-08 08:51:57 +0100108__visible bool kvm_rebooting;
Avi Kivityb7c41452010-12-02 17:52:50 +0200109EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300110
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300111static bool largepages_enabled = true;
112
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +0000113bool kvm_is_reserved_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300114{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200115 if (pfn_valid(pfn))
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +0000116 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300117
118 return true;
119}
120
Avi Kivity6aa8b732006-12-10 02:21:36 -0800121/*
122 * Switches to specified vcpu, until a matching vcpu_put()
123 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300124int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800125{
Avi Kivity15ad7142007-07-11 18:17:21 +0300126 int cpu;
127
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300128 if (mutex_lock_killable(&vcpu->mutex))
129 return -EINTR;
Avi Kivity15ad7142007-07-11 18:17:21 +0300130 cpu = get_cpu();
131 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200132 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300133 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300134 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200135}
136
Carsten Otte313a3dc2007-10-11 19:16:52 +0200137void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800138{
Avi Kivity15ad7142007-07-11 18:17:21 +0300139 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200140 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300141 preempt_notifier_unregister(&vcpu->preempt_notifier);
142 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800143 mutex_unlock(&vcpu->mutex);
144}
145
Avi Kivityd9e368d2007-06-07 19:18:30 +0300146static void ack_flush(void *_completed)
147{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300148}
149
Tang Chen445b8232014-09-24 15:57:55 +0800150bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300151{
Avi Kivity597a5f52008-07-20 14:24:22 +0300152 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030153 cpumask_var_t cpus;
154 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300155 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300156
Li Zefan79f55992009-06-15 14:58:26 +0800157 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030158
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800159 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300160 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800161 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300162 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800163
164 /* Set ->requests bit before we read ->mode */
165 smp_mb();
166
167 if (cpus != NULL && cpu != -1 && cpu != me &&
168 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030169 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300170 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030171 if (unlikely(cpus == NULL))
172 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
173 else if (!cpumask_empty(cpus))
174 smp_call_function_many(cpus, ack_flush, NULL, 1);
175 else
176 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800177 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030178 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030179 return called;
180}
181
Mario Smarducha6d51012015-01-15 15:58:52 -0800182#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
Rusty Russell49846892008-12-08 20:26:24 +1030183void kvm_flush_remote_tlbs(struct kvm *kvm)
184{
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800185 long dirty_count = kvm->tlbs_dirty;
186
187 smp_mb();
Tang Chen445b8232014-09-24 15:57:55 +0800188 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
Rusty Russell49846892008-12-08 20:26:24 +1030189 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800190 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300191}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530192EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Mario Smarducha6d51012015-01-15 15:58:52 -0800193#endif
Avi Kivityd9e368d2007-06-07 19:18:30 +0300194
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500195void kvm_reload_remote_mmus(struct kvm *kvm)
196{
Tang Chen445b8232014-09-24 15:57:55 +0800197 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500198}
199
Marcelo Tosattid8281992012-11-27 23:29:01 -0200200void kvm_make_mclock_inprogress_request(struct kvm *kvm)
201{
Tang Chen445b8232014-09-24 15:57:55 +0800202 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
Marcelo Tosattid8281992012-11-27 23:29:01 -0200203}
204
Yang Zhang3d81bc72013-04-11 19:25:13 +0800205void kvm_make_scan_ioapic_request(struct kvm *kvm)
Yang Zhangc7c9c562013-01-25 10:18:51 +0800206{
Tang Chen445b8232014-09-24 15:57:55 +0800207 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
Yang Zhangc7c9c562013-01-25 10:18:51 +0800208}
209
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000210int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
211{
212 struct page *page;
213 int r;
214
215 mutex_init(&vcpu->mutex);
216 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000217 vcpu->kvm = kvm;
218 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500219 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300220 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200221 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000222
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
224 if (!page) {
225 r = -ENOMEM;
226 goto fail;
227 }
228 vcpu->run = page_address(page);
229
Raghavendra K T4c088492012-07-18 19:07:46 +0530230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530232 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530233
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800234 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000235 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800236 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000237 return 0;
238
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000239fail_free_run:
240 free_page((unsigned long)vcpu->run);
241fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000242 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000243}
244EXPORT_SYMBOL_GPL(kvm_vcpu_init);
245
246void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
247{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500248 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800249 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000250 free_page((unsigned long)vcpu->run);
251}
252EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
253
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200254#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
256{
257 return container_of(mn, struct kvm, mmu_notifier);
258}
259
260static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
263{
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200265 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200266
267 /*
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
275 * before returning.
276 *
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
279 *
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
284 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200285 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200286 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900287
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200288 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200290 /* we've to flush the tlb before the pages can be freed */
291 if (need_tlb_flush)
292 kvm_flush_remote_tlbs(kvm);
293
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900294 spin_unlock(&kvm->mmu_lock);
Tang Chenfe715572014-09-24 15:57:57 +0800295
296 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
297
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900298 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200299}
300
Izik Eidus3da0dd42009-09-23 21:47:18 +0300301static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
304 pte_t pte)
305{
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200307 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300308
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200309 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200314 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300315}
316
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200317static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
319 unsigned long start,
320 unsigned long end)
321{
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200323 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200324
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200325 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200326 spin_lock(&kvm->mmu_lock);
327 /*
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
331 */
332 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800334 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200335 /* we've to flush the tlb before the pages can be freed */
336 if (need_tlb_flush)
337 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900338
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200341}
342
343static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
345 unsigned long start,
346 unsigned long end)
347{
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349
350 spin_lock(&kvm->mmu_lock);
351 /*
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
354 * been freed.
355 */
356 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000357 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200358 /*
359 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200362 */
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
365
366 BUG_ON(kvm->mmu_notifier_count < 0);
367}
368
369static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700371 unsigned long start,
372 unsigned long end)
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200373{
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200375 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200376
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200377 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200378 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200379
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700380 young = kvm_age_hva(kvm, start, end);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200381 if (young)
382 kvm_flush_remote_tlbs(kvm);
383
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900384 spin_unlock(&kvm->mmu_lock);
385 srcu_read_unlock(&kvm->srcu, idx);
386
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200387 return young;
388}
389
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800390static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long address)
393{
394 struct kvm *kvm = mmu_notifier_to_kvm(mn);
395 int young, idx;
396
397 idx = srcu_read_lock(&kvm->srcu);
398 spin_lock(&kvm->mmu_lock);
399 young = kvm_test_age_hva(kvm, address);
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
402
403 return young;
404}
405
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100406static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407 struct mm_struct *mm)
408{
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800410 int idx;
411
412 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300413 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800414 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100415}
416
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200417static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418 .invalidate_page = kvm_mmu_notifier_invalidate_page,
419 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
421 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800422 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300423 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100424 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200425};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200426
427static int kvm_init_mmu_notifier(struct kvm *kvm)
428{
429 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431}
432
433#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434
435static int kvm_init_mmu_notifier(struct kvm *kvm)
436{
437 return 0;
438}
439
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200440#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800442static void kvm_init_memslots_id(struct kvm *kvm)
443{
444 int i;
445 struct kvm_memslots *slots = kvm->memslots;
446
447 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800448 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800449}
450
Carsten Ottee08b9632012-01-04 10:25:20 +0100451static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800452{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100453 int r, i;
454 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800455
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100456 if (!kvm)
457 return ERR_PTR(-ENOMEM);
458
Carsten Ottee08b9632012-01-04 10:25:20 +0100459 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100460 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100461 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200462
463 r = hardware_enable_all();
464 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100465 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200466
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200467#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300468 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200469#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800470
Alex Williamson1e702d92012-12-10 10:33:32 -0700471 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
472
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200473 r = -ENOMEM;
Igor Mammedov74496132015-03-20 12:21:37 +0000474 kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots));
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200475 if (!kvm->memslots)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100476 goto out_err_no_srcu;
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200477
478 /*
479 * Init kvm generation close to the maximum to easily test the
480 * code of handling generation number wrap-around.
481 */
482 kvm->memslots->generation = -150;
483
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800484 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200485 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100486 goto out_err_no_srcu;
487 if (init_srcu_struct(&kvm->irq_srcu))
488 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200489 for (i = 0; i < KVM_NR_BUSES; i++) {
490 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
491 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100492 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200493 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200494 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700495
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900496 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200497 kvm->mm = current->mm;
498 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400499 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800500 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300501 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200502 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300503 atomic_set(&kvm->users_count, 1);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000504 INIT_LIST_HEAD(&kvm->devices);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700505
506 r = kvm_init_mmu_notifier(kvm);
507 if (r)
508 goto out_err;
509
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200510 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000511 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200512 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100513
Avi Kivityf17abe92007-02-21 19:28:04 +0200514 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200515
516out_err:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100517 cleanup_srcu_struct(&kvm->irq_srcu);
518out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100519 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100520out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200521 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100522out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200523 for (i = 0; i < KVM_NR_BUSES; i++)
524 kfree(kvm->buses[i]);
Igor Mammedov74496132015-03-20 12:21:37 +0000525 kvfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100526 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200527 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200528}
529
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900530/*
531 * Avoid using vmalloc for a small buffer.
532 * Should not be used when the size is statically known.
533 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900534void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900535{
536 if (size > PAGE_SIZE)
537 return vzalloc(size);
538 else
539 return kzalloc(size, GFP_KERNEL);
540}
541
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900542void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900543{
544 if (is_vmalloc_addr(addr))
545 vfree(addr);
546 else
547 kfree(addr);
548}
549
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900550static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
551{
552 if (!memslot->dirty_bitmap)
553 return;
554
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900555 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900556 memslot->dirty_bitmap = NULL;
557}
558
Avi Kivity6aa8b732006-12-10 02:21:36 -0800559/*
560 * Free any memory in @free but not in @dont.
561 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530562static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563 struct kvm_memory_slot *dont)
564{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800565 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900566 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800567
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530568 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300569
Avi Kivity6aa8b732006-12-10 02:21:36 -0800570 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800571}
572
Stephen Hemminger79408762013-12-29 12:12:29 -0800573static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800574{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200575 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800576 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800577
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800578 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530579 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200580
Igor Mammedov74496132015-03-20 12:21:37 +0000581 kvfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800582}
583
Scott Wood07f0a7b2013-04-25 14:11:23 +0000584static void kvm_destroy_devices(struct kvm *kvm)
585{
586 struct list_head *node, *tmp;
587
588 list_for_each_safe(node, tmp, &kvm->devices) {
589 struct kvm_device *dev =
590 list_entry(node, struct kvm_device, vm_node);
591
592 list_del(node);
593 dev->ops->destroy(dev);
594 }
595}
596
Avi Kivityf17abe92007-02-21 19:28:04 +0200597static void kvm_destroy_vm(struct kvm *kvm)
598{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200599 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200600 struct mm_struct *mm = kvm->mm;
601
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800602 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200603 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800604 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200605 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200606 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200607 for (i = 0; i < KVM_NR_BUSES; i++)
608 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200609 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200610#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
611 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200612#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300613 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200614#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800615 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000616 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100617 kvm_free_physmem(kvm);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200618 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100619 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 Haskins721eecbf2009-05-20 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 Yoshikawa515a0122010-10-27 18:23:54 +0900655 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900656
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900657 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900658 if (!memslot->dirty_bitmap)
659 return -ENOMEM;
660
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900661 return 0;
662}
663
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800664/*
Igor Mammedov0e60b072014-12-01 17:29:26 +0000665 * Insert memslot and re-sort memslots based on their GFN,
666 * so binary search could be used to lookup GFN.
667 * Sorting algorithm takes advantage of having initially
668 * sorted array and known changed memslot position.
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800669 */
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100670static void update_memslots(struct kvm_memslots *slots,
671 struct kvm_memory_slot *new)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800672{
Paolo Bonzini85931762014-11-14 10:22:07 +0100673 int id = new->id;
674 int i = slots->id_to_index[id];
Igor Mammedov063584d2014-11-13 23:00:13 +0000675 struct kvm_memory_slot *mslots = slots->memslots;
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800676
Paolo Bonzini85931762014-11-14 10:22:07 +0100677 WARN_ON(mslots[i].id != id);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000678 if (!new->npages) {
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100679 WARN_ON(!mslots[i].npages);
Igor Mammedov0e60b072014-12-01 17:29:26 +0000680 new->base_gfn = 0;
Tiejun Chenb0165f12015-01-09 16:29:14 +0800681 new->flags = 0;
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000682 if (mslots[i].npages)
683 slots->used_slots--;
684 } else {
685 if (!mslots[i].npages)
686 slots->used_slots++;
687 }
Igor Mammedov0e60b072014-12-01 17:29:26 +0000688
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000689 while (i < KVM_MEM_SLOTS_NUM - 1 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000690 new->base_gfn <= mslots[i + 1].base_gfn) {
691 if (!mslots[i + 1].npages)
692 break;
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000693 mslots[i] = mslots[i + 1];
694 slots->id_to_index[mslots[i].id] = i;
695 i++;
696 }
Paolo Bonziniefbeec72014-12-27 18:01:00 +0100697
698 /*
699 * The ">=" is needed when creating a slot with base_gfn == 0,
700 * so that it moves before all those with base_gfn == npages == 0.
701 *
702 * On the other hand, if new->npages is zero, the above loop has
703 * already left i pointing to the beginning of the empty part of
704 * mslots, and the ">=" would move the hole backwards in this
705 * case---which is wrong. So skip the loop when deleting a slot.
706 */
707 if (new->npages) {
708 while (i > 0 &&
709 new->base_gfn >= mslots[i - 1].base_gfn) {
710 mslots[i] = mslots[i - 1];
711 slots->id_to_index[mslots[i].id] = i;
712 i--;
713 }
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100714 } else
715 WARN_ON_ONCE(i != slots->used_slots);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800716
Paolo Bonzini85931762014-11-14 10:22:07 +0100717 mslots[i] = *new;
718 slots->id_to_index[mslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800719}
720
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800721static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
722{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800723 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
724
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200725#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800726 valid_flags |= KVM_MEM_READONLY;
727#endif
728
729 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800730 return -EINVAL;
731
732 return 0;
733}
734
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200735static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100736 struct kvm_memslots *slots)
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200737{
738 struct kvm_memslots *old_memslots = kvm->memslots;
739
David Matlackee3d1572014-08-18 15:46:06 -0700740 /*
741 * Set the low bit in the generation, which disables SPTE caching
742 * until the end of synchronize_srcu_expedited.
743 */
744 WARN_ON(old_memslots->generation & 1);
745 slots->generation = old_memslots->generation + 1;
746
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200747 rcu_assign_pointer(kvm->memslots, slots);
748 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900749
David Matlackee3d1572014-08-18 15:46:06 -0700750 /*
751 * Increment the new memslot generation a second time. This prevents
752 * vm exits that race with memslot updates from caching a memslot
753 * generation that will (potentially) be valid forever.
754 */
755 slots->generation++;
756
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900757 kvm_arch_memslots_updated(kvm);
758
759 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200760}
761
Avi Kivity6aa8b732006-12-10 02:21:36 -0800762/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800763 * Allocate some memory and give it an address in the guest physical address
764 * space.
765 *
766 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800767 *
Dominik Dingel02d5d552014-10-27 16:22:56 +0100768 * Must be called holding kvm->slots_lock for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800769 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800770int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900771 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800772{
Avi Kivity8234b222010-12-27 12:08:45 +0200773 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800774 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200775 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900776 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800777 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700778 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900779 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800780
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800781 r = check_memory_region_flags(mem);
782 if (r)
783 goto out;
784
Avi Kivity6aa8b732006-12-10 02:21:36 -0800785 r = -EINVAL;
786 /* General sanity checks */
787 if (mem->memory_size & (PAGE_SIZE - 1))
788 goto out;
789 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
790 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900791 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900792 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900793 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200794 !access_ok(VERIFY_WRITE,
795 (void __user *)(unsigned long)mem->userspace_addr,
796 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600797 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800798 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800799 goto out;
800 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
801 goto out;
802
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900803 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800804 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
805 npages = mem->memory_size >> PAGE_SHIFT;
806
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900807 if (npages > KVM_MEM_MAX_NR_PAGES)
808 goto out;
809
Avi Kivity6aa8b732006-12-10 02:21:36 -0800810 if (!npages)
811 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
812
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900813 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800814
Avi Kivitye36d96f2010-06-21 10:56:36 +0300815 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800816 new.base_gfn = base_gfn;
817 new.npages = npages;
818 new.flags = mem->flags;
819
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900820 if (npages) {
821 if (!old.npages)
822 change = KVM_MR_CREATE;
823 else { /* Modify an existing slot. */
824 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900825 (npages != old.npages) ||
826 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900827 goto out;
828
829 if (base_gfn != old.base_gfn)
830 change = KVM_MR_MOVE;
831 else if (new.flags != old.flags)
832 change = KVM_MR_FLAGS_ONLY;
833 else { /* Nothing to change. */
834 r = 0;
835 goto out;
836 }
837 }
838 } else if (old.npages) {
839 change = KVM_MR_DELETE;
840 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900841 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800842
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900843 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900844 /* Check for overlaps */
845 r = -EEXIST;
846 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900847 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
848 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900849 continue;
850 if (!((base_gfn + npages <= slot->base_gfn) ||
851 (base_gfn >= slot->base_gfn + slot->npages)))
852 goto out;
853 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800854 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800855
Avi Kivity6aa8b732006-12-10 02:21:36 -0800856 /* Free page dirty bitmap if unneeded */
857 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000858 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800859
860 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900861 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900862 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900863
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530864 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900865 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800866 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200867
Avi Kivity6aa8b732006-12-10 02:21:36 -0800868 /* Allocate page dirty bitmap if needed */
869 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900870 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800871 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800872 }
873
Igor Mammedov74496132015-03-20 12:21:37 +0000874 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100875 if (!slots)
876 goto out_free;
Igor Mammedov74496132015-03-20 12:21:37 +0000877 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100878
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900879 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Xiao Guangrong28a37542011-11-24 19:04:35 +0800880 slot = id_to_memslot(slots, mem->slot);
881 slot->flags |= KVM_MEMSLOT_INVALID;
882
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100883 old_memslots = install_new_memslots(kvm, slots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200884
Alex Williamsone40f1932012-12-10 10:32:57 -0700885 /* slot was deleted or moved, clear iommu mapping */
886 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300887 /* From this point no new shadow pages pointing to a deleted,
888 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200889 *
890 * validation of sp->gfn happens in:
891 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
892 * - kvm_is_visible_gfn (mmu_check_roots)
893 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300894 kvm_arch_flush_shadow_memslot(kvm, slot);
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100895
896 /*
897 * We can re-use the old_memslots from above, the only difference
898 * from the currently installed memslots is the invalid flag. This
899 * will get overwritten by update_memslots anyway.
900 */
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700901 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200902 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300903
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900904 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200905 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700906 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200907
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200908 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900909 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200910 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900911 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200912 }
913
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100914 update_memslots(slots, &new);
915 old_memslots = install_new_memslots(kvm, slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900917 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800918
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530919 kvm_free_physmem_slot(kvm, &old, &new);
Igor Mammedov74496132015-03-20 12:21:37 +0000920 kvfree(old_memslots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200921
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200922 /*
923 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800924 * un-mapped and re-mapped if their base changes. Since base change
925 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800926 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800927 * slots (size changes, userspace addr changes and read-only flag
928 * changes) is disallowed above, so any other attribute changes getting
929 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200930 */
931 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800932 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800933 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800934 }
935
Avi Kivity6aa8b732006-12-10 02:21:36 -0800936 return 0;
937
938out_slots:
Igor Mammedov74496132015-03-20 12:21:37 +0000939 kvfree(slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800940out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530941 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800942out:
943 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200944}
945EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800946
947int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900948 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800949{
950 int r;
951
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200952 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900953 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200954 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800955 return r;
956}
Izik Eidus210c7c42007-10-24 23:52:57 +0200957EXPORT_SYMBOL_GPL(kvm_set_memory_region);
958
Stephen Hemminger79408762013-12-29 12:12:29 -0800959static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
960 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200961{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700962 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200963 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900964 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965}
966
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800967int kvm_get_dirty_log(struct kvm *kvm,
968 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969{
970 struct kvm_memory_slot *memslot;
971 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900972 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800973 unsigned long any = 0;
974
Avi Kivity6aa8b732006-12-10 02:21:36 -0800975 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700976 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977 goto out;
978
Xiao Guangrong28a37542011-11-24 19:04:35 +0800979 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800980 r = -ENOENT;
981 if (!memslot->dirty_bitmap)
982 goto out;
983
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900984 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800985
Uri Lublincd1a4a92007-02-22 16:43:09 +0200986 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800987 any = memslot->dirty_bitmap[i];
988
989 r = -EFAULT;
990 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
991 goto out;
992
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800993 if (any)
994 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800995
996 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800997out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800998 return r;
999}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301000EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001001
Mario Smarduchba0513b2015-01-15 15:58:53 -08001002#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1003/**
1004 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1005 * are dirty write protect them for next write.
1006 * @kvm: pointer to kvm instance
1007 * @log: slot id and address to which we copy the log
1008 * @is_dirty: flag set if any page is dirty
1009 *
1010 * We need to keep it in mind that VCPU threads can write to the bitmap
1011 * concurrently. So, to avoid losing track of dirty pages we keep the
1012 * following order:
1013 *
1014 * 1. Take a snapshot of the bit and clear it if needed.
1015 * 2. Write protect the corresponding page.
1016 * 3. Copy the snapshot to the userspace.
1017 * 4. Upon return caller flushes TLB's if needed.
1018 *
1019 * Between 2 and 4, the guest may write to the page using the remaining TLB
1020 * entry. This is not a problem because the page is reported dirty using
1021 * the snapshot taken before and step 4 ensures that writes done after
1022 * exiting to userspace will be logged for the next call.
1023 *
1024 */
1025int kvm_get_dirty_log_protect(struct kvm *kvm,
1026 struct kvm_dirty_log *log, bool *is_dirty)
1027{
1028 struct kvm_memory_slot *memslot;
1029 int r, i;
1030 unsigned long n;
1031 unsigned long *dirty_bitmap;
1032 unsigned long *dirty_bitmap_buffer;
1033
1034 r = -EINVAL;
1035 if (log->slot >= KVM_USER_MEM_SLOTS)
1036 goto out;
1037
1038 memslot = id_to_memslot(kvm->memslots, log->slot);
1039
1040 dirty_bitmap = memslot->dirty_bitmap;
1041 r = -ENOENT;
1042 if (!dirty_bitmap)
1043 goto out;
1044
1045 n = kvm_dirty_bitmap_bytes(memslot);
1046
1047 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1048 memset(dirty_bitmap_buffer, 0, n);
1049
1050 spin_lock(&kvm->mmu_lock);
1051 *is_dirty = false;
1052 for (i = 0; i < n / sizeof(long); i++) {
1053 unsigned long mask;
1054 gfn_t offset;
1055
1056 if (!dirty_bitmap[i])
1057 continue;
1058
1059 *is_dirty = true;
1060
1061 mask = xchg(&dirty_bitmap[i], 0);
1062 dirty_bitmap_buffer[i] = mask;
1063
1064 offset = i * BITS_PER_LONG;
Kai Huang3b0f1d02015-01-28 10:54:23 +08001065 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, offset,
Mario Smarduchba0513b2015-01-15 15:58:53 -08001066 mask);
1067 }
1068
1069 spin_unlock(&kvm->mmu_lock);
1070
1071 r = -EFAULT;
1072 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1073 goto out;
1074
1075 r = 0;
1076out:
1077 return r;
1078}
1079EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1080#endif
1081
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001082bool kvm_largepages_enabled(void)
1083{
1084 return largepages_enabled;
1085}
1086
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001087void kvm_disable_largepages(void)
1088{
1089 largepages_enabled = false;
1090}
1091EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1092
Gleb Natapov49c77542010-10-18 15:22:23 +02001093struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1094{
1095 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1096}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001097EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001098
Izik Eiduse0d62c72007-10-24 23:57:46 +02001099int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1100{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001101 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001102
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001103 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001104 memslot->flags & KVM_MEMSLOT_INVALID)
1105 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001106
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001107 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001108}
1109EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1110
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001111unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1112{
1113 struct vm_area_struct *vma;
1114 unsigned long addr, size;
1115
1116 size = PAGE_SIZE;
1117
1118 addr = gfn_to_hva(kvm, gfn);
1119 if (kvm_is_error_hva(addr))
1120 return PAGE_SIZE;
1121
1122 down_read(&current->mm->mmap_sem);
1123 vma = find_vma(current->mm, addr);
1124 if (!vma)
1125 goto out;
1126
1127 size = vma_kernel_pagesize(vma);
1128
1129out:
1130 up_read(&current->mm->mmap_sem);
1131
1132 return size;
1133}
1134
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001135static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1136{
1137 return slot->flags & KVM_MEM_READONLY;
1138}
1139
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001140static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1141 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001142{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001143 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001144 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001145
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001146 if (memslot_is_readonly(slot) && write)
1147 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001148
1149 if (nr_pages)
1150 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1151
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001152 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001153}
Xiao Guangrong48987782010-08-22 19:11:43 +08001154
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001155static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1156 gfn_t *nr_pages)
1157{
1158 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1159}
1160
1161unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001162 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001163{
1164 return gfn_to_hva_many(slot, gfn, NULL);
1165}
1166EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1167
Xiao Guangrong48987782010-08-22 19:11:43 +08001168unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1169{
Gleb Natapov49c77542010-10-18 15:22:23 +02001170 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001171}
Sheng Yang0d150292008-04-25 21:44:50 +08001172EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001173
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001174/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001175 * If writable is set to false, the hva returned by this function is only
1176 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001177 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001178unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1179 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001180{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001181 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1182
1183 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001184 *writable = !memslot_is_readonly(slot);
1185
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001186 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001187}
1188
Christoffer Dall64d83122014-08-19 12:15:00 +02001189unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1190{
1191 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1192
1193 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1194}
1195
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001196static int kvm_read_hva(void *data, void __user *hva, int len)
1197{
1198 return __copy_from_user(data, hva, len);
1199}
1200
1201static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1202{
1203 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001204}
1205
Geoff Levand39369f72013-04-05 19:20:30 +00001206static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001207 unsigned long start, int write, struct page **page)
1208{
1209 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1210
1211 if (write)
1212 flags |= FOLL_WRITE;
1213
1214 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1215}
1216
Huang Yingfafc3db2011-01-30 11:15:49 +08001217static inline int check_user_page_hwpoison(unsigned long addr)
1218{
1219 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1220
1221 rc = __get_user_pages(current, current->mm, addr, 1,
1222 flags, NULL, NULL, NULL);
1223 return rc == -EHWPOISON;
1224}
1225
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001226/*
1227 * The atomic path to get the writable pfn which will be stored in @pfn,
1228 * true indicates success, otherwise false is returned.
1229 */
1230static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1231 bool write_fault, bool *writable, pfn_t *pfn)
1232{
1233 struct page *page[1];
1234 int npages;
1235
1236 if (!(async || atomic))
1237 return false;
1238
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001239 /*
1240 * Fast pin a writable pfn only if it is a write fault request
1241 * or the caller allows to map a writable pfn for a read fault
1242 * request.
1243 */
1244 if (!(write_fault || writable))
1245 return false;
1246
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001247 npages = __get_user_pages_fast(addr, 1, 1, page);
1248 if (npages == 1) {
1249 *pfn = page_to_pfn(page[0]);
1250
1251 if (writable)
1252 *writable = true;
1253 return true;
1254 }
1255
1256 return false;
1257}
1258
1259/*
1260 * The slow path to get the pfn of the specified host virtual address,
1261 * 1 indicates success, -errno is returned if error is detected.
1262 */
1263static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1264 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001265{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001266 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001267 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001268
1269 might_sleep();
1270
1271 if (writable)
1272 *writable = write_fault;
1273
1274 if (async) {
1275 down_read(&current->mm->mmap_sem);
1276 npages = get_user_page_nowait(current, current->mm,
1277 addr, write_fault, page);
1278 up_read(&current->mm->mmap_sem);
Andrea Arcangeli0664e572015-02-11 15:27:28 -08001279 } else
1280 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1281 write_fault, 0, page,
1282 FOLL_TOUCH|FOLL_HWPOISON);
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001283 if (npages != 1)
1284 return npages;
1285
1286 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001287 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001288 struct page *wpage[1];
1289
1290 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1291 if (npages == 1) {
1292 *writable = true;
1293 put_page(page[0]);
1294 page[0] = wpage[0];
1295 }
1296
1297 npages = 1;
1298 }
1299 *pfn = page_to_pfn(page[0]);
1300 return npages;
1301}
1302
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001303static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1304{
1305 if (unlikely(!(vma->vm_flags & VM_READ)))
1306 return false;
1307
1308 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1309 return false;
1310
1311 return true;
1312}
1313
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001314/*
1315 * Pin guest page in memory and return its pfn.
1316 * @addr: host virtual address which maps memory to the guest
1317 * @atomic: whether this function can sleep
1318 * @async: whether this function need to wait IO complete if the
1319 * host page is not in the memory
1320 * @write_fault: whether we should get a writable host page
1321 * @writable: whether it allows to map a writable host page for !@write_fault
1322 *
1323 * The function will map a writable host page for these two cases:
1324 * 1): @write_fault = true
1325 * 2): @write_fault = false && @writable, @writable will tell the caller
1326 * whether the mapping is writable.
1327 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001328static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1329 bool write_fault, bool *writable)
1330{
1331 struct vm_area_struct *vma;
1332 pfn_t pfn = 0;
1333 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001334
Gleb Natapovaf585b92010-10-14 11:22:46 +02001335 /* we can do it either atomically or asynchronously, not both */
1336 BUG_ON(atomic && async);
1337
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001338 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1339 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001340
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001341 if (atomic)
1342 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001343
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001344 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1345 if (npages == 1)
1346 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001347
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001348 down_read(&current->mm->mmap_sem);
1349 if (npages == -EHWPOISON ||
1350 (!async && check_user_page_hwpoison(addr))) {
1351 pfn = KVM_PFN_ERR_HWPOISON;
1352 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001353 }
Izik Eidus539cb662007-11-11 22:05:04 +02001354
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001355 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001356
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001357 if (vma == NULL)
1358 pfn = KVM_PFN_ERR_FAULT;
1359 else if ((vma->vm_flags & VM_PFNMAP)) {
1360 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1361 vma->vm_pgoff;
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001362 BUG_ON(!kvm_is_reserved_pfn(pfn));
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001363 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001364 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001365 *async = true;
1366 pfn = KVM_PFN_ERR_FAULT;
1367 }
1368exit:
1369 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001370 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001371}
1372
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001373static pfn_t
1374__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1375 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001376{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001377 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1378
1379 if (addr == KVM_HVA_ERR_RO_BAD)
1380 return KVM_PFN_ERR_RO_FAULT;
1381
1382 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001383 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001384
1385 /* Do not map writable pfn in the readonly memslot. */
1386 if (writable && memslot_is_readonly(slot)) {
1387 *writable = false;
1388 writable = NULL;
1389 }
1390
1391 return hva_to_pfn(addr, atomic, async, write_fault,
1392 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001393}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001394
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001395static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1396 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001397{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001398 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001399
Gleb Natapovaf585b92010-10-14 11:22:46 +02001400 if (async)
1401 *async = false;
1402
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001403 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001404
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001405 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1406 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001407}
1408
1409pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1410{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001411 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001412}
1413EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1414
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001415pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1416 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001417{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001418 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001419}
1420EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1421
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001422pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1423{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001424 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001425}
Anthony Liguori35149e22008-04-02 14:46:56 -05001426EXPORT_SYMBOL_GPL(gfn_to_pfn);
1427
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001428pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1429 bool *writable)
1430{
1431 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1432}
1433EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1434
Xiao Guangrongd5661042012-07-17 21:56:16 +08001435pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001436{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001437 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001438}
1439
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001440pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1441{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001442 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001443}
1444EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1445
Xiao Guangrong48987782010-08-22 19:11:43 +08001446int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1447 int nr_pages)
1448{
1449 unsigned long addr;
1450 gfn_t entry;
1451
Gleb Natapov49c77542010-10-18 15:22:23 +02001452 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001453 if (kvm_is_error_hva(addr))
1454 return -1;
1455
1456 if (entry < nr_pages)
1457 return 0;
1458
1459 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1460}
1461EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1462
Xiao Guangronga2766322012-07-26 11:58:59 +08001463static struct page *kvm_pfn_to_page(pfn_t pfn)
1464{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001465 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001466 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001467
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001468 if (kvm_is_reserved_pfn(pfn)) {
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001469 WARN_ON(1);
1470 return KVM_ERR_PTR_BAD_PAGE;
1471 }
1472
Xiao Guangronga2766322012-07-26 11:58:59 +08001473 return pfn_to_page(pfn);
1474}
1475
Anthony Liguori35149e22008-04-02 14:46:56 -05001476struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1477{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001478 pfn_t pfn;
1479
1480 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001481
Xiao Guangronga2766322012-07-26 11:58:59 +08001482 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001483}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001484
Avi Kivity954bbbc2007-03-30 14:02:32 +03001485EXPORT_SYMBOL_GPL(gfn_to_page);
1486
Izik Eidusb4231d62007-11-20 11:49:33 +02001487void kvm_release_page_clean(struct page *page)
1488{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001489 WARN_ON(is_error_page(page));
1490
Anthony Liguori35149e22008-04-02 14:46:56 -05001491 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001492}
1493EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1494
Anthony Liguori35149e22008-04-02 14:46:56 -05001495void kvm_release_pfn_clean(pfn_t pfn)
1496{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001497 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001498 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001499}
1500EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1501
Izik Eidusb4231d62007-11-20 11:49:33 +02001502void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001503{
Xiao Guangronga2766322012-07-26 11:58:59 +08001504 WARN_ON(is_error_page(page));
1505
Anthony Liguori35149e22008-04-02 14:46:56 -05001506 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001507}
Izik Eidusb4231d62007-11-20 11:49:33 +02001508EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001509
Stephen Hemminger79408762013-12-29 12:12:29 -08001510static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001511{
1512 kvm_set_pfn_dirty(pfn);
1513 kvm_release_pfn_clean(pfn);
1514}
Anthony Liguori35149e22008-04-02 14:46:56 -05001515
1516void kvm_set_pfn_dirty(pfn_t pfn)
1517{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001518 if (!kvm_is_reserved_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001519 struct page *page = pfn_to_page(pfn);
1520 if (!PageReserved(page))
1521 SetPageDirty(page);
1522 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001523}
1524EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1525
1526void kvm_set_pfn_accessed(pfn_t pfn)
1527{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001528 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001529 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001530}
1531EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1532
1533void kvm_get_pfn(pfn_t pfn)
1534{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001535 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001536 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001537}
1538EXPORT_SYMBOL_GPL(kvm_get_pfn);
1539
Izik Eidus195aefd2007-10-01 22:14:18 +02001540static int next_segment(unsigned long len, int offset)
1541{
1542 if (len > PAGE_SIZE - offset)
1543 return PAGE_SIZE - offset;
1544 else
1545 return len;
1546}
1547
1548int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1549 int len)
1550{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001551 int r;
1552 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001553
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001554 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001555 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001556 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001557 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001558 if (r)
1559 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001560 return 0;
1561}
1562EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1563
1564int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1565{
1566 gfn_t gfn = gpa >> PAGE_SHIFT;
1567 int seg;
1568 int offset = offset_in_page(gpa);
1569 int ret;
1570
1571 while ((seg = next_segment(len, offset)) != 0) {
1572 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1573 if (ret < 0)
1574 return ret;
1575 offset = 0;
1576 len -= seg;
1577 data += seg;
1578 ++gfn;
1579 }
1580 return 0;
1581}
1582EXPORT_SYMBOL_GPL(kvm_read_guest);
1583
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001584int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1585 unsigned long len)
1586{
1587 int r;
1588 unsigned long addr;
1589 gfn_t gfn = gpa >> PAGE_SHIFT;
1590 int offset = offset_in_page(gpa);
1591
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001592 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001593 if (kvm_is_error_hva(addr))
1594 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001595 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001596 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001597 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001598 if (r)
1599 return -EFAULT;
1600 return 0;
1601}
1602EXPORT_SYMBOL(kvm_read_guest_atomic);
1603
Izik Eidus195aefd2007-10-01 22:14:18 +02001604int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1605 int offset, int len)
1606{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001607 int r;
1608 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001609
Izik Eiduse0506bc2007-11-11 22:10:22 +02001610 addr = gfn_to_hva(kvm, gfn);
1611 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001612 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001613 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001614 if (r)
1615 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001616 mark_page_dirty(kvm, gfn);
1617 return 0;
1618}
1619EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1620
1621int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1622 unsigned long len)
1623{
1624 gfn_t gfn = gpa >> PAGE_SHIFT;
1625 int seg;
1626 int offset = offset_in_page(gpa);
1627 int ret;
1628
1629 while ((seg = next_segment(len, offset)) != 0) {
1630 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1631 if (ret < 0)
1632 return ret;
1633 offset = 0;
1634 len -= seg;
1635 data += seg;
1636 ++gfn;
1637 }
1638 return 0;
1639}
Wincy Vanff651cb2014-12-11 08:52:58 +03001640EXPORT_SYMBOL_GPL(kvm_write_guest);
Izik Eidus195aefd2007-10-01 22:14:18 +02001641
Gleb Natapov49c77542010-10-18 15:22:23 +02001642int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001643 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001644{
1645 struct kvm_memslots *slots = kvm_memslots(kvm);
1646 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001647 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1648 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1649 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1650 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001651
1652 ghc->gpa = gpa;
1653 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001654 ghc->len = len;
1655 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1656 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1657 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001658 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001659 } else {
1660 /*
1661 * If the requested region crosses two memslots, we still
1662 * verify that the entire region is valid here.
1663 */
1664 while (start_gfn <= end_gfn) {
1665 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1666 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1667 &nr_pages_avail);
1668 if (kvm_is_error_hva(ghc->hva))
1669 return -EFAULT;
1670 start_gfn += nr_pages_avail;
1671 }
1672 /* Use the slow path for cross page reads and writes. */
1673 ghc->memslot = NULL;
1674 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001675 return 0;
1676}
1677EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1678
1679int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1680 void *data, unsigned long len)
1681{
1682 struct kvm_memslots *slots = kvm_memslots(kvm);
1683 int r;
1684
Andrew Honig8f964522013-03-29 09:35:21 -07001685 BUG_ON(len > ghc->len);
1686
Gleb Natapov49c77542010-10-18 15:22:23 +02001687 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001688 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1689
1690 if (unlikely(!ghc->memslot))
1691 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001692
1693 if (kvm_is_error_hva(ghc->hva))
1694 return -EFAULT;
1695
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001696 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001697 if (r)
1698 return -EFAULT;
1699 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1700
1701 return 0;
1702}
1703EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1704
Gleb Natapove03b6442011-07-11 15:28:11 -04001705int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1706 void *data, unsigned long len)
1707{
1708 struct kvm_memslots *slots = kvm_memslots(kvm);
1709 int r;
1710
Andrew Honig8f964522013-03-29 09:35:21 -07001711 BUG_ON(len > ghc->len);
1712
Gleb Natapove03b6442011-07-11 15:28:11 -04001713 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001714 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1715
1716 if (unlikely(!ghc->memslot))
1717 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001718
1719 if (kvm_is_error_hva(ghc->hva))
1720 return -EFAULT;
1721
1722 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1723 if (r)
1724 return -EFAULT;
1725
1726 return 0;
1727}
1728EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1729
Izik Eidus195aefd2007-10-01 22:14:18 +02001730int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1731{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001732 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1733
1734 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001735}
1736EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1737
1738int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1739{
1740 gfn_t gfn = gpa >> PAGE_SHIFT;
1741 int seg;
1742 int offset = offset_in_page(gpa);
1743 int ret;
1744
1745 while ((seg = next_segment(len, offset)) != 0) {
1746 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1747 if (ret < 0)
1748 return ret;
1749 offset = 0;
1750 len -= seg;
1751 ++gfn;
1752 }
1753 return 0;
1754}
1755EXPORT_SYMBOL_GPL(kvm_clear_guest);
1756
Stephen Hemminger79408762013-12-29 12:12:29 -08001757static void mark_page_dirty_in_slot(struct kvm *kvm,
1758 struct kvm_memory_slot *memslot,
1759 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001760{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001761 if (memslot && memslot->dirty_bitmap) {
1762 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001763
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001764 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001765 }
1766}
1767
Gleb Natapov49c77542010-10-18 15:22:23 +02001768void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1769{
1770 struct kvm_memory_slot *memslot;
1771
1772 memslot = gfn_to_memslot(kvm, gfn);
1773 mark_page_dirty_in_slot(kvm, memslot, gfn);
1774}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301775EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001776
Paolo Bonzinif7819512015-02-04 18:20:58 +01001777static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1778{
1779 if (kvm_arch_vcpu_runnable(vcpu)) {
1780 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1781 return -EINTR;
1782 }
1783 if (kvm_cpu_has_pending_timer(vcpu))
1784 return -EINTR;
1785 if (signal_pending(current))
1786 return -EINTR;
1787
1788 return 0;
1789}
1790
Eddie Dongb6958ce2007-07-18 12:15:21 +03001791/*
1792 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1793 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001794void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001795{
Paolo Bonzinif7819512015-02-04 18:20:58 +01001796 ktime_t start, cur;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001797 DEFINE_WAIT(wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01001798 bool waited = false;
1799
1800 start = cur = ktime_get();
1801 if (halt_poll_ns) {
1802 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1803 do {
1804 /*
1805 * This sets KVM_REQ_UNHALT if an interrupt
1806 * arrives.
1807 */
1808 if (kvm_vcpu_check_block(vcpu) < 0) {
1809 ++vcpu->stat.halt_successful_poll;
1810 goto out;
1811 }
1812 cur = ktime_get();
1813 } while (single_task_running() && ktime_before(cur, stop));
1814 }
Eddie Dongb6958ce2007-07-18 12:15:21 +03001815
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001816 for (;;) {
1817 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001818
Paolo Bonzinif7819512015-02-04 18:20:58 +01001819 if (kvm_vcpu_check_block(vcpu) < 0)
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001820 break;
1821
Paolo Bonzinif7819512015-02-04 18:20:58 +01001822 waited = true;
Eddie Dongb6958ce2007-07-18 12:15:21 +03001823 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001824 }
1825
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001826 finish_wait(&vcpu->wq, &wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01001827 cur = ktime_get();
1828
1829out:
1830 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001831}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301832EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001833
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001834#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001835/*
1836 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1837 */
1838void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1839{
1840 int me;
1841 int cpu = vcpu->cpu;
1842 wait_queue_head_t *wqp;
1843
1844 wqp = kvm_arch_vcpu_wq(vcpu);
1845 if (waitqueue_active(wqp)) {
1846 wake_up_interruptible(wqp);
1847 ++vcpu->stat.halt_wakeup;
1848 }
1849
1850 me = get_cpu();
1851 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1852 if (kvm_arch_vcpu_should_kick(vcpu))
1853 smp_send_reschedule(cpu);
1854 put_cpu();
1855}
Yang Zhanga20ed542013-04-11 19:25:15 +08001856EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001857#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001858
Dan Carpenterfa933842014-05-23 13:20:42 +03001859int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02001860{
1861 struct pid *pid;
1862 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03001863 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001864
1865 rcu_read_lock();
1866 pid = rcu_dereference(target->pid);
1867 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10001868 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001869 rcu_read_unlock();
1870 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301871 return ret;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301872 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001873 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301874
1875 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001876}
1877EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1878
Raghavendra K T06e48c52012-07-19 15:17:52 +05301879/*
1880 * Helper that checks whether a VCPU is eligible for directed yield.
1881 * Most eligible candidate to yield is decided by following heuristics:
1882 *
1883 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1884 * (preempted lock holder), indicated by @in_spin_loop.
1885 * Set at the beiginning and cleared at the end of interception/PLE handler.
1886 *
1887 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1888 * chance last time (mostly it has become eligible now since we have probably
1889 * yielded to lockholder in last iteration. This is done by toggling
1890 * @dy_eligible each time a VCPU checked for eligibility.)
1891 *
1892 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1893 * to preempted lock-holder could result in wrong VCPU selection and CPU
1894 * burning. Giving priority for a potential lock-holder increases lock
1895 * progress.
1896 *
1897 * Since algorithm is based on heuristics, accessing another VCPU data without
1898 * locking does not harm. It may result in trying to yield to same VCPU, fail
1899 * and continue with next VCPU and so on.
1900 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001901static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301902{
Scott Wood4a55dd72014-01-09 18:43:16 -06001903#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301904 bool eligible;
1905
1906 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02001907 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301908
1909 if (vcpu->spin_loop.in_spin_loop)
1910 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1911
1912 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001913#else
1914 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301915#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001916}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301917
Rik van Riel217ece62011-02-01 09:53:28 -05001918void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001919{
Rik van Riel217ece62011-02-01 09:53:28 -05001920 struct kvm *kvm = me->kvm;
1921 struct kvm_vcpu *vcpu;
1922 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1923 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301924 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001925 int pass;
1926 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001927
Raghavendra K T4c088492012-07-18 19:07:46 +05301928 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001929 /*
1930 * We boost the priority of a VCPU that is runnable but not
1931 * currently running, because it got preempted by something
1932 * else and called schedule in __vcpu_run. Hopefully that
1933 * VCPU is holding the lock that we need and will release it.
1934 * We approximate round-robin by starting at the last boosted VCPU.
1935 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301936 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001937 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001938 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001939 i = last_boosted_vcpu;
1940 continue;
1941 } else if (pass && i > last_boosted_vcpu)
1942 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301943 if (!ACCESS_ONCE(vcpu->preempted))
1944 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001945 if (vcpu == me)
1946 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001947 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001948 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301949 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1950 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301951
1952 yielded = kvm_vcpu_yield_to(vcpu);
1953 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001954 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001955 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301956 } else if (yielded < 0) {
1957 try--;
1958 if (!try)
1959 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001960 }
Rik van Riel217ece62011-02-01 09:53:28 -05001961 }
1962 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301963 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301964
1965 /* Ensure vcpu is not eligible during next spinloop */
1966 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001967}
1968EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1969
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001970static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001971{
1972 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001973 struct page *page;
1974
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001975 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001976 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001977#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001978 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001979 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001980#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001981#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1982 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1983 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1984#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001985 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001986 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001987 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001988 vmf->page = page;
1989 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001990}
1991
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001992static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001993 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001994};
1995
1996static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1997{
1998 vma->vm_ops = &kvm_vcpu_vm_ops;
1999 return 0;
2000}
2001
Avi Kivitybccf2152007-02-21 18:04:26 +02002002static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2003{
2004 struct kvm_vcpu *vcpu = filp->private_data;
2005
Al Viro66c0b392008-04-19 20:33:56 +01002006 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002007 return 0;
2008}
2009
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002010static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02002011 .release = kvm_vcpu_release,
2012 .unlocked_ioctl = kvm_vcpu_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002013#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002014 .compat_ioctl = kvm_vcpu_compat_ioctl,
2015#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002016 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002017 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02002018};
2019
2020/*
2021 * Allocates an inode for the vcpu.
2022 */
2023static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2024{
Yann Droneaud24009b02013-08-24 22:14:07 +02002025 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02002026}
2027
Avi Kivityc5ea7662007-02-20 18:41:05 +02002028/*
2029 * Creates some virtual cpus. Good luck creating more than one.
2030 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002031static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02002032{
2033 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002034 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002035
Andy Honig338c7db2013-11-18 16:09:22 -08002036 if (id >= KVM_MAX_VCPUS)
2037 return -EINVAL;
2038
Gleb Natapov73880c82009-06-09 15:56:28 +03002039 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002040 if (IS_ERR(vcpu))
2041 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002042
Avi Kivity15ad7142007-07-11 18:17:21 +03002043 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2044
Avi Kivity26e52152007-11-20 15:30:24 +02002045 r = kvm_arch_vcpu_setup(vcpu);
2046 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02002047 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02002048
Shaohua Li11ec2802007-07-23 14:51:37 +08002049 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02002050 if (!kvm_vcpu_compatible(vcpu)) {
2051 r = -EINVAL;
2052 goto unlock_vcpu_destroy;
2053 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002054 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2055 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02002056 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002057 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002058
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002059 kvm_for_each_vcpu(r, v, kvm)
2060 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03002061 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02002062 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002063 }
2064
2065 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002066
2067 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01002068 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002069 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03002070 if (r < 0) {
2071 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02002072 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002073 }
2074
2075 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2076 smp_wmb();
2077 atomic_inc(&kvm->online_vcpus);
2078
Gleb Natapov73880c82009-06-09 15:56:28 +03002079 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02002080 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02002081 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002082
Jan Kiszkad7805922011-05-23 10:33:05 +02002083unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03002084 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02002085vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06002086 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002087 return r;
2088}
2089
Avi Kivity1961d272007-03-05 19:46:05 +02002090static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2091{
2092 if (sigset) {
2093 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2094 vcpu->sigset_active = 1;
2095 vcpu->sigset = *sigset;
2096 } else
2097 vcpu->sigset_active = 0;
2098 return 0;
2099}
2100
Avi Kivitybccf2152007-02-21 18:04:26 +02002101static long kvm_vcpu_ioctl(struct file *filp,
2102 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002103{
Avi Kivitybccf2152007-02-21 18:04:26 +02002104 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002105 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002106 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002107 struct kvm_fpu *fpu = NULL;
2108 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002109
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002110 if (vcpu->kvm->mm != current->mm)
2111 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002112
David Matlack2ea75be2014-09-19 16:03:25 -07002113 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2114 return -EINVAL;
2115
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002116#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002117 /*
2118 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2119 * so vcpu_load() would break it.
2120 */
2121 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2122 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2123#endif
2124
2125
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002126 r = vcpu_load(vcpu);
2127 if (r)
2128 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002129 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002130 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002131 r = -EINVAL;
2132 if (arg)
2133 goto out;
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002134 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2135 /* The thread running this VCPU changed. */
2136 struct pid *oldpid = vcpu->pid;
2137 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2138 rcu_assign_pointer(vcpu->pid, newpid);
2139 if (oldpid)
2140 synchronize_rcu();
2141 put_pid(oldpid);
2142 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002143 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002144 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002145 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002146 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002147 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002148
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002149 r = -ENOMEM;
2150 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2151 if (!kvm_regs)
2152 goto out;
2153 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002154 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002155 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002156 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002157 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2158 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002159 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002160out_free1:
2161 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002162 break;
2163 }
2164 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002165 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002166
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002167 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002168 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2169 if (IS_ERR(kvm_regs)) {
2170 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002171 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002172 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002173 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002174 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002175 break;
2176 }
2177 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002178 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2179 r = -ENOMEM;
2180 if (!kvm_sregs)
2181 goto out;
2182 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002183 if (r)
2184 goto out;
2185 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002186 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002187 goto out;
2188 r = 0;
2189 break;
2190 }
2191 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002192 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2193 if (IS_ERR(kvm_sregs)) {
2194 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002195 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002196 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002197 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002198 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002199 break;
2200 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002201 case KVM_GET_MP_STATE: {
2202 struct kvm_mp_state mp_state;
2203
2204 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2205 if (r)
2206 goto out;
2207 r = -EFAULT;
2208 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2209 goto out;
2210 r = 0;
2211 break;
2212 }
2213 case KVM_SET_MP_STATE: {
2214 struct kvm_mp_state mp_state;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2218 goto out;
2219 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002220 break;
2221 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002222 case KVM_TRANSLATE: {
2223 struct kvm_translation tr;
2224
2225 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002226 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002227 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002228 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002229 if (r)
2230 goto out;
2231 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002232 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002233 goto out;
2234 r = 0;
2235 break;
2236 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002237 case KVM_SET_GUEST_DEBUG: {
2238 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002239
2240 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002241 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002242 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002243 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002244 break;
2245 }
Avi Kivity1961d272007-03-05 19:46:05 +02002246 case KVM_SET_SIGNAL_MASK: {
2247 struct kvm_signal_mask __user *sigmask_arg = argp;
2248 struct kvm_signal_mask kvm_sigmask;
2249 sigset_t sigset, *p;
2250
2251 p = NULL;
2252 if (argp) {
2253 r = -EFAULT;
2254 if (copy_from_user(&kvm_sigmask, argp,
2255 sizeof kvm_sigmask))
2256 goto out;
2257 r = -EINVAL;
2258 if (kvm_sigmask.len != sizeof sigset)
2259 goto out;
2260 r = -EFAULT;
2261 if (copy_from_user(&sigset, sigmask_arg->sigset,
2262 sizeof sigset))
2263 goto out;
2264 p = &sigset;
2265 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002266 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002267 break;
2268 }
Avi Kivityb8836732007-04-01 16:34:31 +03002269 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002270 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2271 r = -ENOMEM;
2272 if (!fpu)
2273 goto out;
2274 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002275 if (r)
2276 goto out;
2277 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002278 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002279 goto out;
2280 r = 0;
2281 break;
2282 }
2283 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002284 fpu = memdup_user(argp, sizeof(*fpu));
2285 if (IS_ERR(fpu)) {
2286 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002287 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002288 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002289 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002290 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002291 break;
2292 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002293 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002294 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002295 }
2296out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002297 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002298 kfree(fpu);
2299 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002300 return r;
2301}
2302
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002303#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002304static long kvm_vcpu_compat_ioctl(struct file *filp,
2305 unsigned int ioctl, unsigned long arg)
2306{
2307 struct kvm_vcpu *vcpu = filp->private_data;
2308 void __user *argp = compat_ptr(arg);
2309 int r;
2310
2311 if (vcpu->kvm->mm != current->mm)
2312 return -EIO;
2313
2314 switch (ioctl) {
2315 case KVM_SET_SIGNAL_MASK: {
2316 struct kvm_signal_mask __user *sigmask_arg = argp;
2317 struct kvm_signal_mask kvm_sigmask;
2318 compat_sigset_t csigset;
2319 sigset_t sigset;
2320
2321 if (argp) {
2322 r = -EFAULT;
2323 if (copy_from_user(&kvm_sigmask, argp,
2324 sizeof kvm_sigmask))
2325 goto out;
2326 r = -EINVAL;
2327 if (kvm_sigmask.len != sizeof csigset)
2328 goto out;
2329 r = -EFAULT;
2330 if (copy_from_user(&csigset, sigmask_arg->sigset,
2331 sizeof csigset))
2332 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002333 sigset_from_compat(&sigset, &csigset);
2334 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2335 } else
2336 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002337 break;
2338 }
2339 default:
2340 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2341 }
2342
2343out:
2344 return r;
2345}
2346#endif
2347
Scott Wood852b6d52013-04-12 14:08:42 +00002348static int kvm_device_ioctl_attr(struct kvm_device *dev,
2349 int (*accessor)(struct kvm_device *dev,
2350 struct kvm_device_attr *attr),
2351 unsigned long arg)
2352{
2353 struct kvm_device_attr attr;
2354
2355 if (!accessor)
2356 return -EPERM;
2357
2358 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2359 return -EFAULT;
2360
2361 return accessor(dev, &attr);
2362}
2363
2364static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2365 unsigned long arg)
2366{
2367 struct kvm_device *dev = filp->private_data;
2368
2369 switch (ioctl) {
2370 case KVM_SET_DEVICE_ATTR:
2371 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2372 case KVM_GET_DEVICE_ATTR:
2373 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2374 case KVM_HAS_DEVICE_ATTR:
2375 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2376 default:
2377 if (dev->ops->ioctl)
2378 return dev->ops->ioctl(dev, ioctl, arg);
2379
2380 return -ENOTTY;
2381 }
2382}
2383
Scott Wood852b6d52013-04-12 14:08:42 +00002384static int kvm_device_release(struct inode *inode, struct file *filp)
2385{
2386 struct kvm_device *dev = filp->private_data;
2387 struct kvm *kvm = dev->kvm;
2388
Scott Wood852b6d52013-04-12 14:08:42 +00002389 kvm_put_kvm(kvm);
2390 return 0;
2391}
2392
2393static const struct file_operations kvm_device_fops = {
2394 .unlocked_ioctl = kvm_device_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002395#ifdef CONFIG_KVM_COMPAT
Scott Wooddb6ae612013-04-30 20:00:45 -05002396 .compat_ioctl = kvm_device_ioctl,
2397#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002398 .release = kvm_device_release,
2399};
2400
2401struct kvm_device *kvm_device_from_filp(struct file *filp)
2402{
2403 if (filp->f_op != &kvm_device_fops)
2404 return NULL;
2405
2406 return filp->private_data;
2407}
2408
Will Deacond60eacb2014-09-02 10:27:33 +01002409static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2410#ifdef CONFIG_KVM_MPIC
2411 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2412 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2413#endif
2414
2415#ifdef CONFIG_KVM_XICS
2416 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2417#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002418};
2419
2420int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2421{
2422 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2423 return -ENOSPC;
2424
2425 if (kvm_device_ops_table[type] != NULL)
2426 return -EEXIST;
2427
2428 kvm_device_ops_table[type] = ops;
2429 return 0;
2430}
2431
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002432void kvm_unregister_device_ops(u32 type)
2433{
2434 if (kvm_device_ops_table[type] != NULL)
2435 kvm_device_ops_table[type] = NULL;
2436}
2437
Scott Wood852b6d52013-04-12 14:08:42 +00002438static int kvm_ioctl_create_device(struct kvm *kvm,
2439 struct kvm_create_device *cd)
2440{
2441 struct kvm_device_ops *ops = NULL;
2442 struct kvm_device *dev;
2443 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2444 int ret;
2445
Will Deacond60eacb2014-09-02 10:27:33 +01002446 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002447 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002448
2449 ops = kvm_device_ops_table[cd->type];
2450 if (ops == NULL)
2451 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002452
2453 if (test)
2454 return 0;
2455
2456 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2457 if (!dev)
2458 return -ENOMEM;
2459
2460 dev->ops = ops;
2461 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002462
2463 ret = ops->create(dev, cd->type);
2464 if (ret < 0) {
2465 kfree(dev);
2466 return ret;
2467 }
2468
Yann Droneaud24009b02013-08-24 22:14:07 +02002469 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002470 if (ret < 0) {
2471 ops->destroy(dev);
2472 return ret;
2473 }
2474
Scott Wood07f0a7b2013-04-25 14:11:23 +00002475 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002476 kvm_get_kvm(kvm);
2477 cd->fd = ret;
2478 return 0;
2479}
2480
Alexander Graf92b591a2014-07-14 18:33:08 +02002481static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2482{
2483 switch (arg) {
2484 case KVM_CAP_USER_MEMORY:
2485 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2486 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2487#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2488 case KVM_CAP_SET_BOOT_CPU_ID:
2489#endif
2490 case KVM_CAP_INTERNAL_ERROR_DATA:
2491#ifdef CONFIG_HAVE_KVM_MSI
2492 case KVM_CAP_SIGNAL_MSI:
2493#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002494#ifdef CONFIG_HAVE_KVM_IRQFD
Paolo Bonzinidc9be0f2015-03-05 11:54:46 +01002495 case KVM_CAP_IRQFD:
Alexander Graf92b591a2014-07-14 18:33:08 +02002496 case KVM_CAP_IRQFD_RESAMPLE:
2497#endif
2498 case KVM_CAP_CHECK_EXTENSION_VM:
2499 return 1;
2500#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2501 case KVM_CAP_IRQ_ROUTING:
2502 return KVM_MAX_IRQ_ROUTES;
2503#endif
2504 default:
2505 break;
2506 }
2507 return kvm_vm_ioctl_check_extension(kvm, arg);
2508}
2509
Avi Kivitybccf2152007-02-21 18:04:26 +02002510static long kvm_vm_ioctl(struct file *filp,
2511 unsigned int ioctl, unsigned long arg)
2512{
2513 struct kvm *kvm = filp->private_data;
2514 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002515 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002516
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002517 if (kvm->mm != current->mm)
2518 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002519 switch (ioctl) {
2520 case KVM_CREATE_VCPU:
2521 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002522 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002523 case KVM_SET_USER_MEMORY_REGION: {
2524 struct kvm_userspace_memory_region kvm_userspace_mem;
2525
2526 r = -EFAULT;
2527 if (copy_from_user(&kvm_userspace_mem, argp,
2528 sizeof kvm_userspace_mem))
2529 goto out;
2530
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002531 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002532 break;
2533 }
2534 case KVM_GET_DIRTY_LOG: {
2535 struct kvm_dirty_log log;
2536
2537 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002538 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002539 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002540 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002541 break;
2542 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002543#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2544 case KVM_REGISTER_COALESCED_MMIO: {
2545 struct kvm_coalesced_mmio_zone zone;
2546 r = -EFAULT;
2547 if (copy_from_user(&zone, argp, sizeof zone))
2548 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002549 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002550 break;
2551 }
2552 case KVM_UNREGISTER_COALESCED_MMIO: {
2553 struct kvm_coalesced_mmio_zone zone;
2554 r = -EFAULT;
2555 if (copy_from_user(&zone, argp, sizeof zone))
2556 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002557 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002558 break;
2559 }
2560#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002561 case KVM_IRQFD: {
2562 struct kvm_irqfd data;
2563
2564 r = -EFAULT;
2565 if (copy_from_user(&data, argp, sizeof data))
2566 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002567 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002568 break;
2569 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002570 case KVM_IOEVENTFD: {
2571 struct kvm_ioeventfd data;
2572
2573 r = -EFAULT;
2574 if (copy_from_user(&data, argp, sizeof data))
2575 goto out;
2576 r = kvm_ioeventfd(kvm, &data);
2577 break;
2578 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002579#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2580 case KVM_SET_BOOT_CPU_ID:
2581 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002582 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002583 if (atomic_read(&kvm->online_vcpus) != 0)
2584 r = -EBUSY;
2585 else
2586 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002587 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002588 break;
2589#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002590#ifdef CONFIG_HAVE_KVM_MSI
2591 case KVM_SIGNAL_MSI: {
2592 struct kvm_msi msi;
2593
2594 r = -EFAULT;
2595 if (copy_from_user(&msi, argp, sizeof msi))
2596 goto out;
2597 r = kvm_send_userspace_msi(kvm, &msi);
2598 break;
2599 }
2600#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002601#ifdef __KVM_HAVE_IRQ_LINE
2602 case KVM_IRQ_LINE_STATUS:
2603 case KVM_IRQ_LINE: {
2604 struct kvm_irq_level irq_event;
2605
2606 r = -EFAULT;
2607 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2608 goto out;
2609
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002610 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2611 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002612 if (r)
2613 goto out;
2614
2615 r = -EFAULT;
2616 if (ioctl == KVM_IRQ_LINE_STATUS) {
2617 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2618 goto out;
2619 }
2620
2621 r = 0;
2622 break;
2623 }
2624#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002625#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2626 case KVM_SET_GSI_ROUTING: {
2627 struct kvm_irq_routing routing;
2628 struct kvm_irq_routing __user *urouting;
2629 struct kvm_irq_routing_entry *entries;
2630
2631 r = -EFAULT;
2632 if (copy_from_user(&routing, argp, sizeof(routing)))
2633 goto out;
2634 r = -EINVAL;
2635 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2636 goto out;
2637 if (routing.flags)
2638 goto out;
2639 r = -ENOMEM;
2640 entries = vmalloc(routing.nr * sizeof(*entries));
2641 if (!entries)
2642 goto out;
2643 r = -EFAULT;
2644 urouting = argp;
2645 if (copy_from_user(entries, urouting->entries,
2646 routing.nr * sizeof(*entries)))
2647 goto out_free_irq_routing;
2648 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2649 routing.flags);
2650 out_free_irq_routing:
2651 vfree(entries);
2652 break;
2653 }
2654#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002655 case KVM_CREATE_DEVICE: {
2656 struct kvm_create_device cd;
2657
2658 r = -EFAULT;
2659 if (copy_from_user(&cd, argp, sizeof(cd)))
2660 goto out;
2661
2662 r = kvm_ioctl_create_device(kvm, &cd);
2663 if (r)
2664 goto out;
2665
2666 r = -EFAULT;
2667 if (copy_to_user(argp, &cd, sizeof(cd)))
2668 goto out;
2669
2670 r = 0;
2671 break;
2672 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002673 case KVM_CHECK_EXTENSION:
2674 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2675 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002676 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002677 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002678 }
2679out:
2680 return r;
2681}
2682
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002683#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002684struct compat_kvm_dirty_log {
2685 __u32 slot;
2686 __u32 padding1;
2687 union {
2688 compat_uptr_t dirty_bitmap; /* one bit per page */
2689 __u64 padding2;
2690 };
2691};
2692
2693static long kvm_vm_compat_ioctl(struct file *filp,
2694 unsigned int ioctl, unsigned long arg)
2695{
2696 struct kvm *kvm = filp->private_data;
2697 int r;
2698
2699 if (kvm->mm != current->mm)
2700 return -EIO;
2701 switch (ioctl) {
2702 case KVM_GET_DIRTY_LOG: {
2703 struct compat_kvm_dirty_log compat_log;
2704 struct kvm_dirty_log log;
2705
2706 r = -EFAULT;
2707 if (copy_from_user(&compat_log, (void __user *)arg,
2708 sizeof(compat_log)))
2709 goto out;
2710 log.slot = compat_log.slot;
2711 log.padding1 = compat_log.padding1;
2712 log.padding2 = compat_log.padding2;
2713 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2714
2715 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002716 break;
2717 }
2718 default:
2719 r = kvm_vm_ioctl(filp, ioctl, arg);
2720 }
2721
2722out:
2723 return r;
2724}
2725#endif
2726
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002727static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002728 .release = kvm_vm_release,
2729 .unlocked_ioctl = kvm_vm_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002730#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002731 .compat_ioctl = kvm_vm_compat_ioctl,
2732#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002733 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002734};
2735
Carsten Ottee08b9632012-01-04 10:25:20 +01002736static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002737{
Heiko Carstensaac87632010-10-27 17:22:10 +02002738 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002739 struct kvm *kvm;
2740
Carsten Ottee08b9632012-01-04 10:25:20 +01002741 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002742 if (IS_ERR(kvm))
2743 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002744#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2745 r = kvm_coalesced_mmio_init(kvm);
2746 if (r < 0) {
2747 kvm_put_kvm(kvm);
2748 return r;
2749 }
2750#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002751 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002752 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002753 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002754
Heiko Carstensaac87632010-10-27 17:22:10 +02002755 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002756}
2757
2758static long kvm_dev_ioctl(struct file *filp,
2759 unsigned int ioctl, unsigned long arg)
2760{
Avi Kivity07c45a32007-03-07 13:05:38 +02002761 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002762
2763 switch (ioctl) {
2764 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002765 if (arg)
2766 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002767 r = KVM_API_VERSION;
2768 break;
2769 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002770 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002771 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002772 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02002773 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002774 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002775 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02002776 if (arg)
2777 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002778 r = PAGE_SIZE; /* struct kvm_run */
2779#ifdef CONFIG_X86
2780 r += PAGE_SIZE; /* pio data page */
2781#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002782#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2783 r += PAGE_SIZE; /* coalesced mmio ring page */
2784#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002785 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002786 case KVM_TRACE_ENABLE:
2787 case KVM_TRACE_PAUSE:
2788 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002789 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002790 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002791 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002792 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002793 }
2794out:
2795 return r;
2796}
2797
Avi Kivity6aa8b732006-12-10 02:21:36 -08002798static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002799 .unlocked_ioctl = kvm_dev_ioctl,
2800 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002801 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002802};
2803
2804static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002805 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002806 "kvm",
2807 &kvm_chardev_ops,
2808};
2809
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002810static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002811{
2812 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002813 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002814
Rusty Russell7f59f492008-12-07 21:25:45 +10302815 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002816 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002817
Rusty Russell7f59f492008-12-07 21:25:45 +10302818 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002819
Radim Krčmář13a34e02014-08-28 15:13:03 +02002820 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02002821
2822 if (r) {
2823 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2824 atomic_inc(&hardware_enable_failed);
2825 printk(KERN_INFO "kvm: enabling virtualization on "
2826 "CPU%d failed\n", cpu);
2827 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002828}
2829
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002830static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002831{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002832 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002833 if (kvm_usage_count)
2834 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002835 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002836}
2837
2838static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002839{
2840 int cpu = raw_smp_processor_id();
2841
Rusty Russell7f59f492008-12-07 21:25:45 +10302842 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002843 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302844 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02002845 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03002846}
2847
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002848static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002849{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002850 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002851 if (kvm_usage_count)
2852 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002853 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002854}
2855
Alexander Graf10474ae2009-09-15 11:37:46 +02002856static void hardware_disable_all_nolock(void)
2857{
2858 BUG_ON(!kvm_usage_count);
2859
2860 kvm_usage_count--;
2861 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002862 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002863}
2864
2865static void hardware_disable_all(void)
2866{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002867 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002868 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002869 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002870}
2871
2872static int hardware_enable_all(void)
2873{
2874 int r = 0;
2875
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002876 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002877
2878 kvm_usage_count++;
2879 if (kvm_usage_count == 1) {
2880 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002881 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002882
2883 if (atomic_read(&hardware_enable_failed)) {
2884 hardware_disable_all_nolock();
2885 r = -EBUSY;
2886 }
2887 }
2888
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002889 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002890
2891 return r;
2892}
2893
Avi Kivity774c47f2007-02-12 00:54:47 -08002894static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2895 void *v)
2896{
2897 int cpu = (long)v;
2898
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002899 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002900 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002901 case CPU_DYING:
Avi Kivity6ec8a852007-08-19 15:57:26 +03002902 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2903 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002904 hardware_disable();
Avi Kivity6ec8a852007-08-19 15:57:26 +03002905 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002906 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002907 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2908 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002909 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002910 break;
2911 }
2912 return NOTIFY_OK;
2913}
2914
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002915static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002916 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002917{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002918 /*
2919 * Some (well, at least mine) BIOSes hang on reboot if
2920 * in vmx root mode.
2921 *
2922 * And Intel TXT required VMX off for all cpu when system shutdown.
2923 */
2924 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2925 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002926 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002927 return NOTIFY_OK;
2928}
2929
2930static struct notifier_block kvm_reboot_notifier = {
2931 .notifier_call = kvm_reboot,
2932 .priority = 0,
2933};
2934
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002935static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002936{
2937 int i;
2938
2939 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002940 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002941
2942 kvm_iodevice_destructor(pos);
2943 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002944 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002945}
2946
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002947static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2948 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002949{
Sasha Levin743eeb02011-07-27 16:00:48 +03002950 if (r1->addr < r2->addr)
2951 return -1;
2952 if (r1->addr + r1->len > r2->addr + r2->len)
2953 return 1;
2954 return 0;
2955}
2956
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002957static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2958{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002959 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002960}
2961
Geoff Levand39369f72013-04-05 19:20:30 +00002962static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002963 gpa_t addr, int len)
2964{
Sasha Levin743eeb02011-07-27 16:00:48 +03002965 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2966 .addr = addr,
2967 .len = len,
2968 .dev = dev,
2969 };
2970
2971 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2972 kvm_io_bus_sort_cmp, NULL);
2973
2974 return 0;
2975}
2976
Geoff Levand39369f72013-04-05 19:20:30 +00002977static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002978 gpa_t addr, int len)
2979{
2980 struct kvm_io_range *range, key;
2981 int off;
2982
2983 key = (struct kvm_io_range) {
2984 .addr = addr,
2985 .len = len,
2986 };
2987
2988 range = bsearch(&key, bus->range, bus->dev_count,
2989 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2990 if (range == NULL)
2991 return -ENOENT;
2992
2993 off = range - bus->range;
2994
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002995 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002996 off--;
2997
2998 return off;
2999}
3000
Cornelia Huck126a5af2013-07-03 16:30:53 +02003001static int __kvm_io_bus_write(struct kvm_io_bus *bus,
3002 struct kvm_io_range *range, const void *val)
3003{
3004 int idx;
3005
3006 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3007 if (idx < 0)
3008 return -EOPNOTSUPP;
3009
3010 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003011 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02003012 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
3013 range->len, val))
3014 return idx;
3015 idx++;
3016 }
3017
3018 return -EOPNOTSUPP;
3019}
3020
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003021/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003022int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003023 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003024{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003025 struct kvm_io_bus *bus;
3026 struct kvm_io_range range;
3027 int r;
3028
3029 range = (struct kvm_io_range) {
3030 .addr = addr,
3031 .len = len,
3032 };
3033
3034 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3035 r = __kvm_io_bus_write(bus, &range, val);
3036 return r < 0 ? r : 0;
3037}
3038
3039/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3040int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3041 int len, const void *val, long cookie)
3042{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003043 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03003044 struct kvm_io_range range;
3045
3046 range = (struct kvm_io_range) {
3047 .addr = addr,
3048 .len = len,
3049 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003050
3051 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003052
3053 /* First try the device referenced by cookie. */
3054 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003055 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Cornelia Huck126a5af2013-07-03 16:30:53 +02003056 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
3057 val))
3058 return cookie;
3059
3060 /*
3061 * cookie contained garbage; fall back to search and return the
3062 * correct cookie value.
3063 */
3064 return __kvm_io_bus_write(bus, &range, val);
3065}
3066
3067static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
3068 void *val)
3069{
3070 int idx;
3071
3072 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03003073 if (idx < 0)
3074 return -EOPNOTSUPP;
3075
3076 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003077 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Cornelia Huck126a5af2013-07-03 16:30:53 +02003078 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3079 range->len, val))
3080 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03003081 idx++;
3082 }
3083
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003084 return -EOPNOTSUPP;
3085}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03003086EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003087
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003088/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003089int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3090 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003091{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003092 struct kvm_io_bus *bus;
3093 struct kvm_io_range range;
3094 int r;
3095
3096 range = (struct kvm_io_range) {
3097 .addr = addr,
3098 .len = len,
3099 };
3100
3101 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3102 r = __kvm_io_bus_read(bus, &range, val);
3103 return r < 0 ? r : 0;
3104}
3105
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003106
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003107/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003108int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3109 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003110{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003111 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003112
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003113 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08003114 /* exclude ioeventfd which is limited by maximum fd */
3115 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003116 return -ENOSPC;
3117
Amos Konga13007162012-03-09 12:17:32 +08003118 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3119 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003120 if (!new_bus)
3121 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003122 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3123 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003124 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003125 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3126 synchronize_srcu_expedited(&kvm->srcu);
3127 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003128
3129 return 0;
3130}
3131
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003132/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003133int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3134 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003135{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003136 int i, r;
3137 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003138
Sasha Levincdfca7b2011-12-04 19:36:28 +02003139 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003140 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003141 for (i = 0; i < bus->dev_count; i++)
3142 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003143 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003144 break;
3145 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003146
Amos Konga13007162012-03-09 12:17:32 +08003147 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003148 return r;
Amos Konga13007162012-03-09 12:17:32 +08003149
3150 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3151 sizeof(struct kvm_io_range)), GFP_KERNEL);
3152 if (!new_bus)
3153 return -ENOMEM;
3154
3155 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3156 new_bus->dev_count--;
3157 memcpy(new_bus->range + i, bus->range + i + 1,
3158 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003159
3160 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3161 synchronize_srcu_expedited(&kvm->srcu);
3162 kfree(bus);
3163 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003164}
3165
Avi Kivity774c47f2007-02-12 00:54:47 -08003166static struct notifier_block kvm_cpu_notifier = {
3167 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003168};
3169
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003170static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003171{
3172 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003173 struct kvm *kvm;
3174
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003175 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003176 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003177 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003178 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003179 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003180 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003181}
3182
3183DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3184
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003185static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003186{
3187 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003188 struct kvm *kvm;
3189 struct kvm_vcpu *vcpu;
3190 int i;
3191
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003192 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003193 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003194 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003195 kvm_for_each_vcpu(i, vcpu, kvm)
3196 *val += *(u32 *)((void *)vcpu + offset);
3197
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003198 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003199 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003200}
3201
Avi Kivityba1389b2007-11-18 16:24:12 +02003202DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3203
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003204static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003205 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3206 [KVM_STAT_VM] = &vm_stat_fops,
3207};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003208
Hamo4f69b682011-12-15 14:23:16 +08003209static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003210{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003211 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003212 struct kvm_stats_debugfs_item *p;
3213
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003214 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003215 if (kvm_debugfs_dir == NULL)
3216 goto out;
3217
3218 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003219 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003220 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003221 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003222 if (p->dentry == NULL)
3223 goto out_dir;
3224 }
3225
3226 return 0;
3227
3228out_dir:
3229 debugfs_remove_recursive(kvm_debugfs_dir);
3230out:
3231 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003232}
3233
3234static void kvm_exit_debug(void)
3235{
3236 struct kvm_stats_debugfs_item *p;
3237
3238 for (p = debugfs_entries; p->name; ++p)
3239 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003240 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003241}
3242
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003243static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003244{
Alexander Graf10474ae2009-09-15 11:37:46 +02003245 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003246 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003247 return 0;
3248}
3249
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003250static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003251{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003252 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003253 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003254 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003255 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003256}
3257
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003258static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003259 .suspend = kvm_suspend,
3260 .resume = kvm_resume,
3261};
3262
Avi Kivity15ad7142007-07-11 18:17:21 +03003263static inline
3264struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3265{
3266 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3267}
3268
3269static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3270{
3271 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303272 if (vcpu->preempted)
3273 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003274
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003275 kvm_arch_sched_in(vcpu, cpu);
3276
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003277 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003278}
3279
3280static void kvm_sched_out(struct preempt_notifier *pn,
3281 struct task_struct *next)
3282{
3283 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3284
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303285 if (current->state == TASK_RUNNING)
3286 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003287 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003288}
3289
Avi Kivity0ee75be2010-04-28 15:39:01 +03003290int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003291 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003292{
3293 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003294 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003295
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003296 r = kvm_arch_init(opaque);
3297 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003298 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003299
Asias He7dac16c2013-05-08 10:57:29 +08003300 /*
3301 * kvm_arch_init makes sure there's at most one caller
3302 * for architectures that support multiple implementations,
3303 * like intel and amd on x86.
3304 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3305 * conflicts in case kvm is already setup for another implementation.
3306 */
3307 r = kvm_irqfd_init();
3308 if (r)
3309 goto out_irqfd;
3310
Avi Kivity8437a612009-06-06 14:52:35 -07003311 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303312 r = -ENOMEM;
3313 goto out_free_0;
3314 }
3315
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003316 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003317 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303318 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003319
Yang, Sheng002c7f72007-07-31 14:23:01 +03003320 for_each_online_cpu(cpu) {
3321 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003322 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003323 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003324 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003325 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003326 }
3327
Avi Kivity774c47f2007-02-12 00:54:47 -08003328 r = register_cpu_notifier(&kvm_cpu_notifier);
3329 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003330 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003331 register_reboot_notifier(&kvm_reboot_notifier);
3332
Rusty Russellc16f8622007-07-30 21:12:19 +10003333 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003334 if (!vcpu_align)
3335 vcpu_align = __alignof__(struct kvm_vcpu);
3336 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003337 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003338 if (!kvm_vcpu_cache) {
3339 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003340 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003341 }
3342
Gleb Natapovaf585b92010-10-14 11:22:46 +02003343 r = kvm_async_pf_init();
3344 if (r)
3345 goto out_free;
3346
Avi Kivity6aa8b732006-12-10 02:21:36 -08003347 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003348 kvm_vm_fops.owner = module;
3349 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003350
3351 r = misc_register(&kvm_dev);
3352 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04003353 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003354 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003355 }
3356
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003357 register_syscore_ops(&kvm_syscore_ops);
3358
Avi Kivity15ad7142007-07-11 18:17:21 +03003359 kvm_preempt_ops.sched_in = kvm_sched_in;
3360 kvm_preempt_ops.sched_out = kvm_sched_out;
3361
Hamo4f69b682011-12-15 14:23:16 +08003362 r = kvm_init_debug();
3363 if (r) {
3364 printk(KERN_ERR "kvm: create debugfs files failed\n");
3365 goto out_undebugfs;
3366 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003367
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003368 r = kvm_vfio_ops_init();
3369 WARN_ON(r);
3370
Avi Kivityc7addb92007-09-16 18:58:32 +02003371 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003372
Hamo4f69b682011-12-15 14:23:16 +08003373out_undebugfs:
3374 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003375 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003376out_unreg:
3377 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003378out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003379 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003380out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003381 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003382 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003383out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003384out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003385 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303386out_free_0a:
3387 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003388out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003389 kvm_irqfd_exit();
3390out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003391 kvm_arch_exit();
3392out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003393 return r;
3394}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003395EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003396
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003397void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003398{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003399 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003400 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003401 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003402 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003403 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003404 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003405 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003406 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003407 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003408 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003409 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303410 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08003411 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003412}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003413EXPORT_SYMBOL_GPL(kvm_exit);