blob: d3fc9399062a5034b99eaa3d12c855699fdbf608 [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
Andre Przywaraaf669ac2015-03-26 14:39:29 +000019#include <kvm/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
Christian Borntraeger0fa97782015-02-27 16:50:10 +010069static unsigned int halt_poll_ns;
Paolo Bonzinif7819512015-02-04 18:20:58 +010070module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
71
Marcelo Tosattifa40a822009-06-04 15:08:24 -030072/*
73 * Ordering of locks:
74 *
Xiubo Lib7d409d2015-02-26 14:58:24 +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;
Xiubo Lif4fee932015-02-26 14:58:21 +080083static int kvm_usage_count;
Alexander Graf10474ae2009-09-15 11:37:46 +020084static 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 Yoshikawaa36a57b12010-10-27 18:22:19 +0900542static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
543{
544 if (!memslot->dirty_bitmap)
545 return;
546
Thomas Huth548ef282015-02-24 21:29:25 +0100547 kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900548 memslot->dirty_bitmap = NULL;
549}
550
Avi Kivity6aa8b732006-12-10 02:21:36 -0800551/*
552 * Free any memory in @free but not in @dont.
553 */
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530554static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
Avi Kivity6aa8b732006-12-10 02:21:36 -0800555 struct kvm_memory_slot *dont)
556{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900558 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800559
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530560 kvm_arch_free_memslot(kvm, free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300561
Avi Kivity6aa8b732006-12-10 02:21:36 -0800562 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563}
564
Stephen Hemminger79408762013-12-29 12:12:29 -0800565static void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200567 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800568 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800570 kvm_for_each_memslot(memslot, slots)
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530571 kvm_free_physmem_slot(kvm, memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200572
Igor Mammedov74496132015-03-20 12:21:37 +0000573 kvfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800574}
575
Scott Wood07f0a7b2013-04-25 14:11:23 +0000576static void kvm_destroy_devices(struct kvm *kvm)
577{
578 struct list_head *node, *tmp;
579
580 list_for_each_safe(node, tmp, &kvm->devices) {
581 struct kvm_device *dev =
582 list_entry(node, struct kvm_device, vm_node);
583
584 list_del(node);
585 dev->ops->destroy(dev);
586 }
587}
588
Avi Kivityf17abe92007-02-21 19:28:04 +0200589static void kvm_destroy_vm(struct kvm *kvm)
590{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200591 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200592 struct mm_struct *mm = kvm->mm;
593
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800594 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200595 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800596 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200597 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200598 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200599 for (i = 0; i < KVM_NR_BUSES; i++)
600 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200601 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200602#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
603 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200604#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300605 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200606#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800607 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000608 kvm_destroy_devices(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100609 kvm_free_physmem(kvm);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200610 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100611 cleanup_srcu_struct(&kvm->srcu);
612 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200613 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200614 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200615}
616
Izik Eidusd39f13b2008-03-30 16:01:25 +0300617void kvm_get_kvm(struct kvm *kvm)
618{
619 atomic_inc(&kvm->users_count);
620}
621EXPORT_SYMBOL_GPL(kvm_get_kvm);
622
623void kvm_put_kvm(struct kvm *kvm)
624{
625 if (atomic_dec_and_test(&kvm->users_count))
626 kvm_destroy_vm(kvm);
627}
628EXPORT_SYMBOL_GPL(kvm_put_kvm);
629
630
Avi Kivityf17abe92007-02-21 19:28:04 +0200631static int kvm_vm_release(struct inode *inode, struct file *filp)
632{
633 struct kvm *kvm = filp->private_data;
634
Gregory Haskins721eecb2009-05-20 10:30:49 -0400635 kvm_irqfd_release(kvm);
636
Izik Eidusd39f13b2008-03-30 16:01:25 +0300637 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800638 return 0;
639}
640
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900641/*
642 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900643 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900644 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900645static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
646{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900647 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900648
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900649 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900650 if (!memslot->dirty_bitmap)
651 return -ENOMEM;
652
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900653 return 0;
654}
655
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800656/*
Igor Mammedov0e60b072014-12-01 17:29:26 +0000657 * Insert memslot and re-sort memslots based on their GFN,
658 * so binary search could be used to lookup GFN.
659 * Sorting algorithm takes advantage of having initially
660 * sorted array and known changed memslot position.
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800661 */
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100662static void update_memslots(struct kvm_memslots *slots,
663 struct kvm_memory_slot *new)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800664{
Paolo Bonzini85931762014-11-14 10:22:07 +0100665 int id = new->id;
666 int i = slots->id_to_index[id];
Igor Mammedov063584d2014-11-13 23:00:13 +0000667 struct kvm_memory_slot *mslots = slots->memslots;
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800668
Paolo Bonzini85931762014-11-14 10:22:07 +0100669 WARN_ON(mslots[i].id != id);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000670 if (!new->npages) {
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100671 WARN_ON(!mslots[i].npages);
Igor Mammedov0e60b072014-12-01 17:29:26 +0000672 new->base_gfn = 0;
Tiejun Chenb0165f12015-01-09 16:29:14 +0800673 new->flags = 0;
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000674 if (mslots[i].npages)
675 slots->used_slots--;
676 } else {
677 if (!mslots[i].npages)
678 slots->used_slots++;
679 }
Igor Mammedov0e60b072014-12-01 17:29:26 +0000680
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000681 while (i < KVM_MEM_SLOTS_NUM - 1 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000682 new->base_gfn <= mslots[i + 1].base_gfn) {
683 if (!mslots[i + 1].npages)
684 break;
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000685 mslots[i] = mslots[i + 1];
686 slots->id_to_index[mslots[i].id] = i;
687 i++;
688 }
Paolo Bonziniefbeec72014-12-27 18:01:00 +0100689
690 /*
691 * The ">=" is needed when creating a slot with base_gfn == 0,
692 * so that it moves before all those with base_gfn == npages == 0.
693 *
694 * On the other hand, if new->npages is zero, the above loop has
695 * already left i pointing to the beginning of the empty part of
696 * mslots, and the ">=" would move the hole backwards in this
697 * case---which is wrong. So skip the loop when deleting a slot.
698 */
699 if (new->npages) {
700 while (i > 0 &&
701 new->base_gfn >= mslots[i - 1].base_gfn) {
702 mslots[i] = mslots[i - 1];
703 slots->id_to_index[mslots[i].id] = i;
704 i--;
705 }
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100706 } else
707 WARN_ON_ONCE(i != slots->used_slots);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800708
Paolo Bonzini85931762014-11-14 10:22:07 +0100709 mslots[i] = *new;
710 slots->id_to_index[mslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800711}
712
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800713static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
714{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800715 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
716
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200717#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800718 valid_flags |= KVM_MEM_READONLY;
719#endif
720
721 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800722 return -EINVAL;
723
724 return 0;
725}
726
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200727static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100728 struct kvm_memslots *slots)
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200729{
730 struct kvm_memslots *old_memslots = kvm->memslots;
731
David Matlackee3d1572014-08-18 15:46:06 -0700732 /*
733 * Set the low bit in the generation, which disables SPTE caching
734 * until the end of synchronize_srcu_expedited.
735 */
736 WARN_ON(old_memslots->generation & 1);
737 slots->generation = old_memslots->generation + 1;
738
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200739 rcu_assign_pointer(kvm->memslots, slots);
740 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900741
David Matlackee3d1572014-08-18 15:46:06 -0700742 /*
743 * Increment the new memslot generation a second time. This prevents
744 * vm exits that race with memslot updates from caching a memslot
745 * generation that will (potentially) be valid forever.
746 */
747 slots->generation++;
748
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900749 kvm_arch_memslots_updated(kvm);
750
751 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200752}
753
Avi Kivity6aa8b732006-12-10 02:21:36 -0800754/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800755 * Allocate some memory and give it an address in the guest physical address
756 * space.
757 *
758 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800759 *
Dominik Dingel02d5d552014-10-27 16:22:56 +0100760 * Must be called holding kvm->slots_lock for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800761 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800762int __kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900763 struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800764{
Avi Kivity8234b222010-12-27 12:08:45 +0200765 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800766 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200767 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900768 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800769 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700770 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900771 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800772
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800773 r = check_memory_region_flags(mem);
774 if (r)
775 goto out;
776
Avi Kivity6aa8b732006-12-10 02:21:36 -0800777 r = -EINVAL;
778 /* General sanity checks */
779 if (mem->memory_size & (PAGE_SIZE - 1))
780 goto out;
781 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
782 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900783 /* We can read the guest memory with __xxx_user() later on. */
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900784 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900785 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200786 !access_ok(VERIFY_WRITE,
787 (void __user *)(unsigned long)mem->userspace_addr,
788 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600789 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800790 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800791 goto out;
792 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
793 goto out;
794
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900795 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800796 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
797 npages = mem->memory_size >> PAGE_SHIFT;
798
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900799 if (npages > KVM_MEM_MAX_NR_PAGES)
800 goto out;
801
Avi Kivity6aa8b732006-12-10 02:21:36 -0800802 if (!npages)
803 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
804
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900805 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800806
Avi Kivitye36d96f2010-06-21 10:56:36 +0300807 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800808 new.base_gfn = base_gfn;
809 new.npages = npages;
810 new.flags = mem->flags;
811
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900812 if (npages) {
813 if (!old.npages)
814 change = KVM_MR_CREATE;
815 else { /* Modify an existing slot. */
816 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900817 (npages != old.npages) ||
818 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900819 goto out;
820
821 if (base_gfn != old.base_gfn)
822 change = KVM_MR_MOVE;
823 else if (new.flags != old.flags)
824 change = KVM_MR_FLAGS_ONLY;
825 else { /* Nothing to change. */
826 r = 0;
827 goto out;
828 }
829 }
830 } else if (old.npages) {
831 change = KVM_MR_DELETE;
832 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900833 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900835 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900836 /* Check for overlaps */
837 r = -EEXIST;
838 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900839 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
840 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900841 continue;
842 if (!((base_gfn + npages <= slot->base_gfn) ||
843 (base_gfn >= slot->base_gfn + slot->npages)))
844 goto out;
845 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800846 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800847
Avi Kivity6aa8b732006-12-10 02:21:36 -0800848 /* Free page dirty bitmap if unneeded */
849 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000850 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800851
852 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900853 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900854 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900855
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530856 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900857 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800858 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200859
Avi Kivity6aa8b732006-12-10 02:21:36 -0800860 /* Allocate page dirty bitmap if needed */
861 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900862 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800863 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800864 }
865
Igor Mammedov74496132015-03-20 12:21:37 +0000866 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100867 if (!slots)
868 goto out_free;
Igor Mammedov74496132015-03-20 12:21:37 +0000869 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100870
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900871 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Xiao Guangrong28a37542011-11-24 19:04:35 +0800872 slot = id_to_memslot(slots, mem->slot);
873 slot->flags |= KVM_MEMSLOT_INVALID;
874
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100875 old_memslots = install_new_memslots(kvm, slots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200876
Alex Williamsone40f1932012-12-10 10:32:57 -0700877 /* slot was deleted or moved, clear iommu mapping */
878 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300879 /* From this point no new shadow pages pointing to a deleted,
880 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200881 *
882 * validation of sp->gfn happens in:
Xiubo Lib7d409d2015-02-26 14:58:24 +0800883 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
884 * - kvm_is_visible_gfn (mmu_check_roots)
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200885 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300886 kvm_arch_flush_shadow_memslot(kvm, slot);
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100887
888 /*
889 * We can re-use the old_memslots from above, the only difference
890 * from the currently installed memslots is the invalid flag. This
891 * will get overwritten by update_memslots anyway.
892 */
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700893 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200894 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300895
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +0900896 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200897 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700898 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200899
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200900 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900901 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200902 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900903 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200904 }
905
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100906 update_memslots(slots, &new);
907 old_memslots = install_new_memslots(kvm, slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800908
Takuya Yoshikawa84826442013-02-27 19:45:25 +0900909 kvm_arch_commit_memory_region(kvm, mem, &old, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800910
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530911 kvm_free_physmem_slot(kvm, &old, &new);
Igor Mammedov74496132015-03-20 12:21:37 +0000912 kvfree(old_memslots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200913
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200914 /*
915 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -0800916 * un-mapped and re-mapped if their base changes. Since base change
917 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +0800918 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -0800919 * slots (size changes, userspace addr changes and read-only flag
920 * changes) is disallowed above, so any other attribute changes getting
921 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +0200922 */
923 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +0800924 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +0800925 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +0800926 }
927
Avi Kivity6aa8b732006-12-10 02:21:36 -0800928 return 0;
929
930out_slots:
Igor Mammedov74496132015-03-20 12:21:37 +0000931 kvfree(slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800932out_free:
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530933 kvm_free_physmem_slot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800934out:
935 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200936}
937EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800938
939int kvm_set_memory_region(struct kvm *kvm,
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900940 struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800941{
942 int r;
943
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200944 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900945 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200946 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800947 return r;
948}
Izik Eidus210c7c42007-10-24 23:52:57 +0200949EXPORT_SYMBOL_GPL(kvm_set_memory_region);
950
Stephen Hemminger79408762013-12-29 12:12:29 -0800951static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
952 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +0200953{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700954 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200955 return -EINVAL;
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +0900956 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957}
958
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800959int kvm_get_dirty_log(struct kvm *kvm,
960 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800961{
962 struct kvm_memory_slot *memslot;
963 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900964 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965 unsigned long any = 0;
966
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700968 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800969 goto out;
970
Xiao Guangrong28a37542011-11-24 19:04:35 +0800971 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800972 r = -ENOENT;
973 if (!memslot->dirty_bitmap)
974 goto out;
975
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900976 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977
Uri Lublincd1a4a92007-02-22 16:43:09 +0200978 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800979 any = memslot->dirty_bitmap[i];
980
981 r = -EFAULT;
982 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
983 goto out;
984
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800985 if (any)
986 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800987
988 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800989out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800990 return r;
991}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530992EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800993
Mario Smarduchba0513b2015-01-15 15:58:53 -0800994#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
995/**
996 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
997 * are dirty write protect them for next write.
998 * @kvm: pointer to kvm instance
999 * @log: slot id and address to which we copy the log
1000 * @is_dirty: flag set if any page is dirty
1001 *
1002 * We need to keep it in mind that VCPU threads can write to the bitmap
1003 * concurrently. So, to avoid losing track of dirty pages we keep the
1004 * following order:
1005 *
1006 * 1. Take a snapshot of the bit and clear it if needed.
1007 * 2. Write protect the corresponding page.
1008 * 3. Copy the snapshot to the userspace.
1009 * 4. Upon return caller flushes TLB's if needed.
1010 *
1011 * Between 2 and 4, the guest may write to the page using the remaining TLB
1012 * entry. This is not a problem because the page is reported dirty using
1013 * the snapshot taken before and step 4 ensures that writes done after
1014 * exiting to userspace will be logged for the next call.
1015 *
1016 */
1017int kvm_get_dirty_log_protect(struct kvm *kvm,
1018 struct kvm_dirty_log *log, bool *is_dirty)
1019{
1020 struct kvm_memory_slot *memslot;
1021 int r, i;
1022 unsigned long n;
1023 unsigned long *dirty_bitmap;
1024 unsigned long *dirty_bitmap_buffer;
1025
1026 r = -EINVAL;
1027 if (log->slot >= KVM_USER_MEM_SLOTS)
1028 goto out;
1029
1030 memslot = id_to_memslot(kvm->memslots, log->slot);
1031
1032 dirty_bitmap = memslot->dirty_bitmap;
1033 r = -ENOENT;
1034 if (!dirty_bitmap)
1035 goto out;
1036
1037 n = kvm_dirty_bitmap_bytes(memslot);
1038
1039 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1040 memset(dirty_bitmap_buffer, 0, n);
1041
1042 spin_lock(&kvm->mmu_lock);
1043 *is_dirty = false;
1044 for (i = 0; i < n / sizeof(long); i++) {
1045 unsigned long mask;
1046 gfn_t offset;
1047
1048 if (!dirty_bitmap[i])
1049 continue;
1050
1051 *is_dirty = true;
1052
1053 mask = xchg(&dirty_bitmap[i], 0);
1054 dirty_bitmap_buffer[i] = mask;
1055
Takuya Yoshikawa58d29302015-03-17 16:19:58 +09001056 if (mask) {
1057 offset = i * BITS_PER_LONG;
1058 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1059 offset, mask);
1060 }
Mario Smarduchba0513b2015-01-15 15:58:53 -08001061 }
1062
1063 spin_unlock(&kvm->mmu_lock);
1064
1065 r = -EFAULT;
1066 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1067 goto out;
1068
1069 r = 0;
1070out:
1071 return r;
1072}
1073EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1074#endif
1075
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001076bool kvm_largepages_enabled(void)
1077{
1078 return largepages_enabled;
1079}
1080
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001081void kvm_disable_largepages(void)
1082{
1083 largepages_enabled = false;
1084}
1085EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1086
Gleb Natapov49c77542010-10-18 15:22:23 +02001087struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1088{
1089 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1090}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001091EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001092
Izik Eiduse0d62c72007-10-24 23:57:46 +02001093int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1094{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001095 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001096
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001097 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001098 memslot->flags & KVM_MEMSLOT_INVALID)
1099 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001100
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001101 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001102}
1103EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1104
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001105unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1106{
1107 struct vm_area_struct *vma;
1108 unsigned long addr, size;
1109
1110 size = PAGE_SIZE;
1111
1112 addr = gfn_to_hva(kvm, gfn);
1113 if (kvm_is_error_hva(addr))
1114 return PAGE_SIZE;
1115
1116 down_read(&current->mm->mmap_sem);
1117 vma = find_vma(current->mm, addr);
1118 if (!vma)
1119 goto out;
1120
1121 size = vma_kernel_pagesize(vma);
1122
1123out:
1124 up_read(&current->mm->mmap_sem);
1125
1126 return size;
1127}
1128
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001129static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1130{
1131 return slot->flags & KVM_MEM_READONLY;
1132}
1133
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001134static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1135 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001136{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001137 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001138 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001139
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001140 if (memslot_is_readonly(slot) && write)
1141 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001142
1143 if (nr_pages)
1144 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1145
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001146 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001147}
Xiao Guangrong48987782010-08-22 19:11:43 +08001148
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001149static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1150 gfn_t *nr_pages)
1151{
1152 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1153}
1154
1155unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001156 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001157{
1158 return gfn_to_hva_many(slot, gfn, NULL);
1159}
1160EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1161
Xiao Guangrong48987782010-08-22 19:11:43 +08001162unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1163{
Gleb Natapov49c77542010-10-18 15:22:23 +02001164 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001165}
Sheng Yang0d150292008-04-25 21:44:50 +08001166EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001167
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001168/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001169 * If writable is set to false, the hva returned by this function is only
1170 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001171 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001172unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1173 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001174{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001175 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1176
1177 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001178 *writable = !memslot_is_readonly(slot);
1179
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001180 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001181}
1182
Christoffer Dall64d83122014-08-19 12:15:00 +02001183unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1184{
1185 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1186
1187 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1188}
1189
Geoff Levand39369f72013-04-05 19:20:30 +00001190static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001191 unsigned long start, int write, struct page **page)
1192{
1193 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1194
1195 if (write)
1196 flags |= FOLL_WRITE;
1197
1198 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1199}
1200
Huang Yingfafc3db2011-01-30 11:15:49 +08001201static inline int check_user_page_hwpoison(unsigned long addr)
1202{
1203 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1204
1205 rc = __get_user_pages(current, current->mm, addr, 1,
1206 flags, NULL, NULL, NULL);
1207 return rc == -EHWPOISON;
1208}
1209
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001210/*
1211 * The atomic path to get the writable pfn which will be stored in @pfn,
1212 * true indicates success, otherwise false is returned.
1213 */
1214static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1215 bool write_fault, bool *writable, pfn_t *pfn)
1216{
1217 struct page *page[1];
1218 int npages;
1219
1220 if (!(async || atomic))
1221 return false;
1222
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001223 /*
1224 * Fast pin a writable pfn only if it is a write fault request
1225 * or the caller allows to map a writable pfn for a read fault
1226 * request.
1227 */
1228 if (!(write_fault || writable))
1229 return false;
1230
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001231 npages = __get_user_pages_fast(addr, 1, 1, page);
1232 if (npages == 1) {
1233 *pfn = page_to_pfn(page[0]);
1234
1235 if (writable)
1236 *writable = true;
1237 return true;
1238 }
1239
1240 return false;
1241}
1242
1243/*
1244 * The slow path to get the pfn of the specified host virtual address,
1245 * 1 indicates success, -errno is returned if error is detected.
1246 */
1247static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1248 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001249{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001250 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001251 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001252
1253 might_sleep();
1254
1255 if (writable)
1256 *writable = write_fault;
1257
1258 if (async) {
1259 down_read(&current->mm->mmap_sem);
1260 npages = get_user_page_nowait(current, current->mm,
1261 addr, write_fault, page);
1262 up_read(&current->mm->mmap_sem);
Andrea Arcangeli0664e572015-02-11 15:27:28 -08001263 } else
1264 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1265 write_fault, 0, page,
1266 FOLL_TOUCH|FOLL_HWPOISON);
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001267 if (npages != 1)
1268 return npages;
1269
1270 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001271 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001272 struct page *wpage[1];
1273
1274 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1275 if (npages == 1) {
1276 *writable = true;
1277 put_page(page[0]);
1278 page[0] = wpage[0];
1279 }
1280
1281 npages = 1;
1282 }
1283 *pfn = page_to_pfn(page[0]);
1284 return npages;
1285}
1286
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001287static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1288{
1289 if (unlikely(!(vma->vm_flags & VM_READ)))
1290 return false;
1291
1292 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1293 return false;
1294
1295 return true;
1296}
1297
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001298/*
1299 * Pin guest page in memory and return its pfn.
1300 * @addr: host virtual address which maps memory to the guest
1301 * @atomic: whether this function can sleep
1302 * @async: whether this function need to wait IO complete if the
1303 * host page is not in the memory
1304 * @write_fault: whether we should get a writable host page
1305 * @writable: whether it allows to map a writable host page for !@write_fault
1306 *
1307 * The function will map a writable host page for these two cases:
1308 * 1): @write_fault = true
1309 * 2): @write_fault = false && @writable, @writable will tell the caller
1310 * whether the mapping is writable.
1311 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001312static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1313 bool write_fault, bool *writable)
1314{
1315 struct vm_area_struct *vma;
1316 pfn_t pfn = 0;
1317 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001318
Gleb Natapovaf585b92010-10-14 11:22:46 +02001319 /* we can do it either atomically or asynchronously, not both */
1320 BUG_ON(atomic && async);
1321
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001322 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1323 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001324
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001325 if (atomic)
1326 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001327
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001328 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1329 if (npages == 1)
1330 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001331
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001332 down_read(&current->mm->mmap_sem);
1333 if (npages == -EHWPOISON ||
1334 (!async && check_user_page_hwpoison(addr))) {
1335 pfn = KVM_PFN_ERR_HWPOISON;
1336 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001337 }
Izik Eidus539cb662007-11-11 22:05:04 +02001338
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001339 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001340
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001341 if (vma == NULL)
1342 pfn = KVM_PFN_ERR_FAULT;
1343 else if ((vma->vm_flags & VM_PFNMAP)) {
1344 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1345 vma->vm_pgoff;
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001346 BUG_ON(!kvm_is_reserved_pfn(pfn));
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001347 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001348 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001349 *async = true;
1350 pfn = KVM_PFN_ERR_FAULT;
1351 }
1352exit:
1353 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001354 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001355}
1356
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001357static pfn_t
1358__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1359 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001360{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001361 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1362
1363 if (addr == KVM_HVA_ERR_RO_BAD)
1364 return KVM_PFN_ERR_RO_FAULT;
1365
1366 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001367 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001368
1369 /* Do not map writable pfn in the readonly memslot. */
1370 if (writable && memslot_is_readonly(slot)) {
1371 *writable = false;
1372 writable = NULL;
1373 }
1374
1375 return hva_to_pfn(addr, atomic, async, write_fault,
1376 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001377}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001378
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001379static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1380 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001381{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001382 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001383
Gleb Natapovaf585b92010-10-14 11:22:46 +02001384 if (async)
1385 *async = false;
1386
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001387 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001388
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001389 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1390 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001391}
1392
1393pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1394{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001395 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001396}
1397EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1398
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001399pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1400 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001401{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001402 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001403}
1404EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1405
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001406pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1407{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001408 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001409}
Anthony Liguori35149e22008-04-02 14:46:56 -05001410EXPORT_SYMBOL_GPL(gfn_to_pfn);
1411
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001412pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1413 bool *writable)
1414{
1415 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1416}
1417EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1418
Xiao Guangrongd5661042012-07-17 21:56:16 +08001419pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001420{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001421 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001422}
1423
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001424pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1425{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001426 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001427}
1428EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1429
Xiao Guangrong48987782010-08-22 19:11:43 +08001430int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1431 int nr_pages)
1432{
1433 unsigned long addr;
1434 gfn_t entry;
1435
Gleb Natapov49c77542010-10-18 15:22:23 +02001436 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001437 if (kvm_is_error_hva(addr))
1438 return -1;
1439
1440 if (entry < nr_pages)
1441 return 0;
1442
1443 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1444}
1445EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1446
Xiao Guangronga2766322012-07-26 11:58:59 +08001447static struct page *kvm_pfn_to_page(pfn_t pfn)
1448{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001449 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001450 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001451
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001452 if (kvm_is_reserved_pfn(pfn)) {
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001453 WARN_ON(1);
1454 return KVM_ERR_PTR_BAD_PAGE;
1455 }
1456
Xiao Guangronga2766322012-07-26 11:58:59 +08001457 return pfn_to_page(pfn);
1458}
1459
Anthony Liguori35149e22008-04-02 14:46:56 -05001460struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1461{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001462 pfn_t pfn;
1463
1464 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001465
Xiao Guangronga2766322012-07-26 11:58:59 +08001466 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001467}
1468EXPORT_SYMBOL_GPL(gfn_to_page);
1469
Izik Eidusb4231d62007-11-20 11:49:33 +02001470void kvm_release_page_clean(struct page *page)
1471{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001472 WARN_ON(is_error_page(page));
1473
Anthony Liguori35149e22008-04-02 14:46:56 -05001474 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001475}
1476EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1477
Anthony Liguori35149e22008-04-02 14:46:56 -05001478void kvm_release_pfn_clean(pfn_t pfn)
1479{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001480 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001481 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001482}
1483EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1484
Izik Eidusb4231d62007-11-20 11:49:33 +02001485void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001486{
Xiao Guangronga2766322012-07-26 11:58:59 +08001487 WARN_ON(is_error_page(page));
1488
Anthony Liguori35149e22008-04-02 14:46:56 -05001489 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001490}
Izik Eidusb4231d62007-11-20 11:49:33 +02001491EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001492
Stephen Hemminger79408762013-12-29 12:12:29 -08001493static void kvm_release_pfn_dirty(pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001494{
1495 kvm_set_pfn_dirty(pfn);
1496 kvm_release_pfn_clean(pfn);
1497}
Anthony Liguori35149e22008-04-02 14:46:56 -05001498
1499void kvm_set_pfn_dirty(pfn_t pfn)
1500{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001501 if (!kvm_is_reserved_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001502 struct page *page = pfn_to_page(pfn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08001503
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001504 if (!PageReserved(page))
1505 SetPageDirty(page);
1506 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001507}
1508EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1509
1510void kvm_set_pfn_accessed(pfn_t pfn)
1511{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001512 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001513 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001514}
1515EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1516
1517void kvm_get_pfn(pfn_t pfn)
1518{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001519 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001520 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001521}
1522EXPORT_SYMBOL_GPL(kvm_get_pfn);
1523
Izik Eidus195aefd2007-10-01 22:14:18 +02001524static int next_segment(unsigned long len, int offset)
1525{
1526 if (len > PAGE_SIZE - offset)
1527 return PAGE_SIZE - offset;
1528 else
1529 return len;
1530}
1531
1532int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1533 int len)
1534{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001535 int r;
1536 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001537
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001538 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001539 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001540 return -EFAULT;
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001541 r = __copy_from_user(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001542 if (r)
1543 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001544 return 0;
1545}
1546EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1547
1548int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1549{
1550 gfn_t gfn = gpa >> PAGE_SHIFT;
1551 int seg;
1552 int offset = offset_in_page(gpa);
1553 int ret;
1554
1555 while ((seg = next_segment(len, offset)) != 0) {
1556 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1557 if (ret < 0)
1558 return ret;
1559 offset = 0;
1560 len -= seg;
1561 data += seg;
1562 ++gfn;
1563 }
1564 return 0;
1565}
1566EXPORT_SYMBOL_GPL(kvm_read_guest);
1567
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001568int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1569 unsigned long len)
1570{
1571 int r;
1572 unsigned long addr;
1573 gfn_t gfn = gpa >> PAGE_SHIFT;
1574 int offset = offset_in_page(gpa);
1575
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001576 addr = gfn_to_hva_prot(kvm, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001577 if (kvm_is_error_hva(addr))
1578 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001579 pagefault_disable();
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001580 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001581 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001582 if (r)
1583 return -EFAULT;
1584 return 0;
1585}
1586EXPORT_SYMBOL(kvm_read_guest_atomic);
1587
Izik Eidus195aefd2007-10-01 22:14:18 +02001588int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1589 int offset, int len)
1590{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001591 int r;
1592 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001593
Izik Eiduse0506bc2007-11-11 22:10:22 +02001594 addr = gfn_to_hva(kvm, gfn);
1595 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001596 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001597 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001598 if (r)
1599 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001600 mark_page_dirty(kvm, gfn);
1601 return 0;
1602}
1603EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1604
1605int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1606 unsigned long len)
1607{
1608 gfn_t gfn = gpa >> PAGE_SHIFT;
1609 int seg;
1610 int offset = offset_in_page(gpa);
1611 int ret;
1612
1613 while ((seg = next_segment(len, offset)) != 0) {
1614 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1615 if (ret < 0)
1616 return ret;
1617 offset = 0;
1618 len -= seg;
1619 data += seg;
1620 ++gfn;
1621 }
1622 return 0;
1623}
Wincy Vanff651cb2014-12-11 08:52:58 +03001624EXPORT_SYMBOL_GPL(kvm_write_guest);
Izik Eidus195aefd2007-10-01 22:14:18 +02001625
Gleb Natapov49c77542010-10-18 15:22:23 +02001626int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001627 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001628{
1629 struct kvm_memslots *slots = kvm_memslots(kvm);
1630 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001631 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1632 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1633 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1634 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001635
1636 ghc->gpa = gpa;
1637 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001638 ghc->len = len;
1639 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
Radim Krčmářca3f0872015-04-08 14:16:48 +02001640 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1641 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001642 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001643 } else {
1644 /*
1645 * If the requested region crosses two memslots, we still
1646 * verify that the entire region is valid here.
1647 */
1648 while (start_gfn <= end_gfn) {
1649 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1650 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1651 &nr_pages_avail);
1652 if (kvm_is_error_hva(ghc->hva))
1653 return -EFAULT;
1654 start_gfn += nr_pages_avail;
1655 }
1656 /* Use the slow path for cross page reads and writes. */
1657 ghc->memslot = NULL;
1658 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001659 return 0;
1660}
1661EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1662
1663int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1664 void *data, unsigned long len)
1665{
1666 struct kvm_memslots *slots = kvm_memslots(kvm);
1667 int r;
1668
Andrew Honig8f964522013-03-29 09:35:21 -07001669 BUG_ON(len > ghc->len);
1670
Gleb Natapov49c77542010-10-18 15:22:23 +02001671 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001672 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1673
1674 if (unlikely(!ghc->memslot))
1675 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001676
1677 if (kvm_is_error_hva(ghc->hva))
1678 return -EFAULT;
1679
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001680 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001681 if (r)
1682 return -EFAULT;
1683 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1684
1685 return 0;
1686}
1687EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1688
Gleb Natapove03b6442011-07-11 15:28:11 -04001689int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1690 void *data, unsigned long len)
1691{
1692 struct kvm_memslots *slots = kvm_memslots(kvm);
1693 int r;
1694
Andrew Honig8f964522013-03-29 09:35:21 -07001695 BUG_ON(len > ghc->len);
1696
Gleb Natapove03b6442011-07-11 15:28:11 -04001697 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001698 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1699
1700 if (unlikely(!ghc->memslot))
1701 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001702
1703 if (kvm_is_error_hva(ghc->hva))
1704 return -EFAULT;
1705
1706 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1707 if (r)
1708 return -EFAULT;
1709
1710 return 0;
1711}
1712EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1713
Izik Eidus195aefd2007-10-01 22:14:18 +02001714int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1715{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001716 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1717
1718 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001719}
1720EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1721
1722int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1723{
1724 gfn_t gfn = gpa >> PAGE_SHIFT;
1725 int seg;
1726 int offset = offset_in_page(gpa);
1727 int ret;
1728
Kevin Mulveybfda0e82015-02-20 08:21:36 -05001729 while ((seg = next_segment(len, offset)) != 0) {
Izik Eidus195aefd2007-10-01 22:14:18 +02001730 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1731 if (ret < 0)
1732 return ret;
1733 offset = 0;
1734 len -= seg;
1735 ++gfn;
1736 }
1737 return 0;
1738}
1739EXPORT_SYMBOL_GPL(kvm_clear_guest);
1740
Stephen Hemminger79408762013-12-29 12:12:29 -08001741static void mark_page_dirty_in_slot(struct kvm *kvm,
1742 struct kvm_memory_slot *memslot,
1743 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001744{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001745 if (memslot && memslot->dirty_bitmap) {
1746 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001747
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001748 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001749 }
1750}
1751
Gleb Natapov49c77542010-10-18 15:22:23 +02001752void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1753{
1754 struct kvm_memory_slot *memslot;
1755
1756 memslot = gfn_to_memslot(kvm, gfn);
1757 mark_page_dirty_in_slot(kvm, memslot, gfn);
1758}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301759EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02001760
Paolo Bonzinif7819512015-02-04 18:20:58 +01001761static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1762{
1763 if (kvm_arch_vcpu_runnable(vcpu)) {
1764 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1765 return -EINTR;
1766 }
1767 if (kvm_cpu_has_pending_timer(vcpu))
1768 return -EINTR;
1769 if (signal_pending(current))
1770 return -EINTR;
1771
1772 return 0;
1773}
1774
Eddie Dongb6958ce2007-07-18 12:15:21 +03001775/*
1776 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1777 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001778void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001779{
Paolo Bonzinif7819512015-02-04 18:20:58 +01001780 ktime_t start, cur;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001781 DEFINE_WAIT(wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01001782 bool waited = false;
1783
1784 start = cur = ktime_get();
1785 if (halt_poll_ns) {
1786 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08001787
Paolo Bonzinif7819512015-02-04 18:20:58 +01001788 do {
1789 /*
1790 * This sets KVM_REQ_UNHALT if an interrupt
1791 * arrives.
1792 */
1793 if (kvm_vcpu_check_block(vcpu) < 0) {
1794 ++vcpu->stat.halt_successful_poll;
1795 goto out;
1796 }
1797 cur = ktime_get();
1798 } while (single_task_running() && ktime_before(cur, stop));
1799 }
Eddie Dongb6958ce2007-07-18 12:15:21 +03001800
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001801 for (;;) {
1802 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001803
Paolo Bonzinif7819512015-02-04 18:20:58 +01001804 if (kvm_vcpu_check_block(vcpu) < 0)
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001805 break;
1806
Paolo Bonzinif7819512015-02-04 18:20:58 +01001807 waited = true;
Eddie Dongb6958ce2007-07-18 12:15:21 +03001808 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001809 }
1810
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001811 finish_wait(&vcpu->wq, &wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01001812 cur = ktime_get();
1813
1814out:
1815 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001816}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301817EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001818
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001819#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001820/*
1821 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1822 */
1823void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1824{
1825 int me;
1826 int cpu = vcpu->cpu;
1827 wait_queue_head_t *wqp;
1828
1829 wqp = kvm_arch_vcpu_wq(vcpu);
1830 if (waitqueue_active(wqp)) {
1831 wake_up_interruptible(wqp);
1832 ++vcpu->stat.halt_wakeup;
1833 }
1834
1835 me = get_cpu();
1836 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1837 if (kvm_arch_vcpu_should_kick(vcpu))
1838 smp_send_reschedule(cpu);
1839 put_cpu();
1840}
Yang Zhanga20ed542013-04-11 19:25:15 +08001841EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001842#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001843
Dan Carpenterfa933842014-05-23 13:20:42 +03001844int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02001845{
1846 struct pid *pid;
1847 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03001848 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001849
1850 rcu_read_lock();
1851 pid = rcu_dereference(target->pid);
1852 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10001853 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001854 rcu_read_unlock();
1855 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301856 return ret;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301857 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001858 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301859
1860 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001861}
1862EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1863
Raghavendra K T06e48c52012-07-19 15:17:52 +05301864/*
1865 * Helper that checks whether a VCPU is eligible for directed yield.
1866 * Most eligible candidate to yield is decided by following heuristics:
1867 *
1868 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1869 * (preempted lock holder), indicated by @in_spin_loop.
1870 * Set at the beiginning and cleared at the end of interception/PLE handler.
1871 *
1872 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1873 * chance last time (mostly it has become eligible now since we have probably
1874 * yielded to lockholder in last iteration. This is done by toggling
1875 * @dy_eligible each time a VCPU checked for eligibility.)
1876 *
1877 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1878 * to preempted lock-holder could result in wrong VCPU selection and CPU
1879 * burning. Giving priority for a potential lock-holder increases lock
1880 * progress.
1881 *
1882 * Since algorithm is based on heuristics, accessing another VCPU data without
1883 * locking does not harm. It may result in trying to yield to same VCPU, fail
1884 * and continue with next VCPU and so on.
1885 */
Stephen Hemminger79408762013-12-29 12:12:29 -08001886static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05301887{
Scott Wood4a55dd72014-01-09 18:43:16 -06001888#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05301889 bool eligible;
1890
1891 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02001892 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301893
1894 if (vcpu->spin_loop.in_spin_loop)
1895 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1896
1897 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06001898#else
1899 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301900#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06001901}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301902
Rik van Riel217ece62011-02-01 09:53:28 -05001903void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001904{
Rik van Riel217ece62011-02-01 09:53:28 -05001905 struct kvm *kvm = me->kvm;
1906 struct kvm_vcpu *vcpu;
1907 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1908 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301909 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001910 int pass;
1911 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001912
Raghavendra K T4c088492012-07-18 19:07:46 +05301913 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001914 /*
1915 * We boost the priority of a VCPU that is runnable but not
1916 * currently running, because it got preempted by something
1917 * else and called schedule in __vcpu_run. Hopefully that
1918 * VCPU is holding the lock that we need and will release it.
1919 * We approximate round-robin by starting at the last boosted VCPU.
1920 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301921 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001922 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001923 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001924 i = last_boosted_vcpu;
1925 continue;
1926 } else if (pass && i > last_boosted_vcpu)
1927 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05301928 if (!ACCESS_ONCE(vcpu->preempted))
1929 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05001930 if (vcpu == me)
1931 continue;
Michael Mueller98f4a142014-02-26 16:14:18 +01001932 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05001933 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301934 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1935 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301936
1937 yielded = kvm_vcpu_yield_to(vcpu);
1938 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001939 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001940 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301941 } else if (yielded < 0) {
1942 try--;
1943 if (!try)
1944 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001945 }
Rik van Riel217ece62011-02-01 09:53:28 -05001946 }
1947 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301948 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301949
1950 /* Ensure vcpu is not eligible during next spinloop */
1951 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001952}
1953EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1954
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001955static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001956{
1957 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001958 struct page *page;
1959
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001960 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001961 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001962#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001963 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001964 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001965#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001966#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1967 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1968 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1969#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001970 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001971 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001972 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001973 vmf->page = page;
1974 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001975}
1976
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001977static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001978 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001979};
1980
1981static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1982{
1983 vma->vm_ops = &kvm_vcpu_vm_ops;
1984 return 0;
1985}
1986
Avi Kivitybccf2152007-02-21 18:04:26 +02001987static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1988{
1989 struct kvm_vcpu *vcpu = filp->private_data;
1990
Al Viro66c0b392008-04-19 20:33:56 +01001991 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001992 return 0;
1993}
1994
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001995static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001996 .release = kvm_vcpu_release,
1997 .unlocked_ioctl = kvm_vcpu_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01001998#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02001999 .compat_ioctl = kvm_vcpu_compat_ioctl,
2000#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002001 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002002 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02002003};
2004
2005/*
2006 * Allocates an inode for the vcpu.
2007 */
2008static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2009{
Yann Droneaud24009b02013-08-24 22:14:07 +02002010 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02002011}
2012
Avi Kivityc5ea7662007-02-20 18:41:05 +02002013/*
2014 * Creates some virtual cpus. Good luck creating more than one.
2015 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002016static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02002017{
2018 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002019 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002020
Andy Honig338c7db2013-11-18 16:09:22 -08002021 if (id >= KVM_MAX_VCPUS)
2022 return -EINVAL;
2023
Gleb Natapov73880c82009-06-09 15:56:28 +03002024 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002025 if (IS_ERR(vcpu))
2026 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002027
Avi Kivity15ad7142007-07-11 18:17:21 +03002028 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2029
Avi Kivity26e52152007-11-20 15:30:24 +02002030 r = kvm_arch_vcpu_setup(vcpu);
2031 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02002032 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02002033
Shaohua Li11ec2802007-07-23 14:51:37 +08002034 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02002035 if (!kvm_vcpu_compatible(vcpu)) {
2036 r = -EINVAL;
2037 goto unlock_vcpu_destroy;
2038 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002039 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2040 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02002041 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002042 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002043
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002044 kvm_for_each_vcpu(r, v, kvm)
2045 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03002046 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02002047 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002048 }
2049
2050 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002051
2052 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01002053 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002054 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03002055 if (r < 0) {
2056 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02002057 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002058 }
2059
2060 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2061 smp_wmb();
2062 atomic_inc(&kvm->online_vcpus);
2063
Gleb Natapov73880c82009-06-09 15:56:28 +03002064 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02002065 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02002066 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002067
Jan Kiszkad7805922011-05-23 10:33:05 +02002068unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03002069 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02002070vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06002071 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002072 return r;
2073}
2074
Avi Kivity1961d272007-03-05 19:46:05 +02002075static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2076{
2077 if (sigset) {
2078 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2079 vcpu->sigset_active = 1;
2080 vcpu->sigset = *sigset;
2081 } else
2082 vcpu->sigset_active = 0;
2083 return 0;
2084}
2085
Avi Kivitybccf2152007-02-21 18:04:26 +02002086static long kvm_vcpu_ioctl(struct file *filp,
2087 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002088{
Avi Kivitybccf2152007-02-21 18:04:26 +02002089 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002090 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002091 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002092 struct kvm_fpu *fpu = NULL;
2093 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002094
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002095 if (vcpu->kvm->mm != current->mm)
2096 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002097
David Matlack2ea75be2014-09-19 16:03:25 -07002098 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2099 return -EINVAL;
2100
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002101#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002102 /*
2103 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2104 * so vcpu_load() would break it.
2105 */
Jens Freimann47b43c52014-11-11 20:57:06 +01002106 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
Avi Kivity2122ff52010-05-13 11:25:04 +03002107 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2108#endif
2109
2110
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002111 r = vcpu_load(vcpu);
2112 if (r)
2113 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002114 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002115 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002116 r = -EINVAL;
2117 if (arg)
2118 goto out;
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002119 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2120 /* The thread running this VCPU changed. */
2121 struct pid *oldpid = vcpu->pid;
2122 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002123
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002124 rcu_assign_pointer(vcpu->pid, newpid);
2125 if (oldpid)
2126 synchronize_rcu();
2127 put_pid(oldpid);
2128 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002129 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002130 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002131 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002132 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002133 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002134
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002135 r = -ENOMEM;
2136 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2137 if (!kvm_regs)
2138 goto out;
2139 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002140 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002141 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002142 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002143 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2144 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002145 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002146out_free1:
2147 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002148 break;
2149 }
2150 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002151 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002152
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002153 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002154 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2155 if (IS_ERR(kvm_regs)) {
2156 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002157 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002158 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002159 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002160 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002161 break;
2162 }
2163 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002164 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2165 r = -ENOMEM;
2166 if (!kvm_sregs)
2167 goto out;
2168 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002169 if (r)
2170 goto out;
2171 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002172 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002173 goto out;
2174 r = 0;
2175 break;
2176 }
2177 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002178 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2179 if (IS_ERR(kvm_sregs)) {
2180 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002181 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002182 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002183 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002184 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002185 break;
2186 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002187 case KVM_GET_MP_STATE: {
2188 struct kvm_mp_state mp_state;
2189
2190 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2191 if (r)
2192 goto out;
2193 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002194 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002195 goto out;
2196 r = 0;
2197 break;
2198 }
2199 case KVM_SET_MP_STATE: {
2200 struct kvm_mp_state mp_state;
2201
2202 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002203 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002204 goto out;
2205 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002206 break;
2207 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002208 case KVM_TRANSLATE: {
2209 struct kvm_translation tr;
2210
2211 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002212 if (copy_from_user(&tr, argp, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002213 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002214 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002215 if (r)
2216 goto out;
2217 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002218 if (copy_to_user(argp, &tr, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002219 goto out;
2220 r = 0;
2221 break;
2222 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002223 case KVM_SET_GUEST_DEBUG: {
2224 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002225
2226 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002227 if (copy_from_user(&dbg, argp, sizeof(dbg)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002228 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002229 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002230 break;
2231 }
Avi Kivity1961d272007-03-05 19:46:05 +02002232 case KVM_SET_SIGNAL_MASK: {
2233 struct kvm_signal_mask __user *sigmask_arg = argp;
2234 struct kvm_signal_mask kvm_sigmask;
2235 sigset_t sigset, *p;
2236
2237 p = NULL;
2238 if (argp) {
2239 r = -EFAULT;
2240 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002241 sizeof(kvm_sigmask)))
Avi Kivity1961d272007-03-05 19:46:05 +02002242 goto out;
2243 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002244 if (kvm_sigmask.len != sizeof(sigset))
Avi Kivity1961d272007-03-05 19:46:05 +02002245 goto out;
2246 r = -EFAULT;
2247 if (copy_from_user(&sigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002248 sizeof(sigset)))
Avi Kivity1961d272007-03-05 19:46:05 +02002249 goto out;
2250 p = &sigset;
2251 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002252 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002253 break;
2254 }
Avi Kivityb8836732007-04-01 16:34:31 +03002255 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002256 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2257 r = -ENOMEM;
2258 if (!fpu)
2259 goto out;
2260 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002261 if (r)
2262 goto out;
2263 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002264 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002265 goto out;
2266 r = 0;
2267 break;
2268 }
2269 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002270 fpu = memdup_user(argp, sizeof(*fpu));
2271 if (IS_ERR(fpu)) {
2272 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002273 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002274 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002275 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002276 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002277 break;
2278 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002279 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002280 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002281 }
2282out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002283 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002284 kfree(fpu);
2285 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002286 return r;
2287}
2288
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002289#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002290static long kvm_vcpu_compat_ioctl(struct file *filp,
2291 unsigned int ioctl, unsigned long arg)
2292{
2293 struct kvm_vcpu *vcpu = filp->private_data;
2294 void __user *argp = compat_ptr(arg);
2295 int r;
2296
2297 if (vcpu->kvm->mm != current->mm)
2298 return -EIO;
2299
2300 switch (ioctl) {
2301 case KVM_SET_SIGNAL_MASK: {
2302 struct kvm_signal_mask __user *sigmask_arg = argp;
2303 struct kvm_signal_mask kvm_sigmask;
2304 compat_sigset_t csigset;
2305 sigset_t sigset;
2306
2307 if (argp) {
2308 r = -EFAULT;
2309 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002310 sizeof(kvm_sigmask)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002311 goto out;
2312 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002313 if (kvm_sigmask.len != sizeof(csigset))
Alexander Graf1dda6062011-06-08 02:45:37 +02002314 goto out;
2315 r = -EFAULT;
2316 if (copy_from_user(&csigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002317 sizeof(csigset)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002318 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002319 sigset_from_compat(&sigset, &csigset);
2320 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2321 } else
2322 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002323 break;
2324 }
2325 default:
2326 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2327 }
2328
2329out:
2330 return r;
2331}
2332#endif
2333
Scott Wood852b6d52013-04-12 14:08:42 +00002334static int kvm_device_ioctl_attr(struct kvm_device *dev,
2335 int (*accessor)(struct kvm_device *dev,
2336 struct kvm_device_attr *attr),
2337 unsigned long arg)
2338{
2339 struct kvm_device_attr attr;
2340
2341 if (!accessor)
2342 return -EPERM;
2343
2344 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2345 return -EFAULT;
2346
2347 return accessor(dev, &attr);
2348}
2349
2350static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2351 unsigned long arg)
2352{
2353 struct kvm_device *dev = filp->private_data;
2354
2355 switch (ioctl) {
2356 case KVM_SET_DEVICE_ATTR:
2357 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2358 case KVM_GET_DEVICE_ATTR:
2359 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2360 case KVM_HAS_DEVICE_ATTR:
2361 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2362 default:
2363 if (dev->ops->ioctl)
2364 return dev->ops->ioctl(dev, ioctl, arg);
2365
2366 return -ENOTTY;
2367 }
2368}
2369
Scott Wood852b6d52013-04-12 14:08:42 +00002370static int kvm_device_release(struct inode *inode, struct file *filp)
2371{
2372 struct kvm_device *dev = filp->private_data;
2373 struct kvm *kvm = dev->kvm;
2374
Scott Wood852b6d52013-04-12 14:08:42 +00002375 kvm_put_kvm(kvm);
2376 return 0;
2377}
2378
2379static const struct file_operations kvm_device_fops = {
2380 .unlocked_ioctl = kvm_device_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002381#ifdef CONFIG_KVM_COMPAT
Scott Wooddb6ae612013-04-30 20:00:45 -05002382 .compat_ioctl = kvm_device_ioctl,
2383#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002384 .release = kvm_device_release,
2385};
2386
2387struct kvm_device *kvm_device_from_filp(struct file *filp)
2388{
2389 if (filp->f_op != &kvm_device_fops)
2390 return NULL;
2391
2392 return filp->private_data;
2393}
2394
Will Deacond60eacb2014-09-02 10:27:33 +01002395static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2396#ifdef CONFIG_KVM_MPIC
2397 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2398 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2399#endif
2400
2401#ifdef CONFIG_KVM_XICS
2402 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2403#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002404};
2405
2406int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2407{
2408 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2409 return -ENOSPC;
2410
2411 if (kvm_device_ops_table[type] != NULL)
2412 return -EEXIST;
2413
2414 kvm_device_ops_table[type] = ops;
2415 return 0;
2416}
2417
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002418void kvm_unregister_device_ops(u32 type)
2419{
2420 if (kvm_device_ops_table[type] != NULL)
2421 kvm_device_ops_table[type] = NULL;
2422}
2423
Scott Wood852b6d52013-04-12 14:08:42 +00002424static int kvm_ioctl_create_device(struct kvm *kvm,
2425 struct kvm_create_device *cd)
2426{
2427 struct kvm_device_ops *ops = NULL;
2428 struct kvm_device *dev;
2429 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2430 int ret;
2431
Will Deacond60eacb2014-09-02 10:27:33 +01002432 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002433 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002434
2435 ops = kvm_device_ops_table[cd->type];
2436 if (ops == NULL)
2437 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002438
2439 if (test)
2440 return 0;
2441
2442 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2443 if (!dev)
2444 return -ENOMEM;
2445
2446 dev->ops = ops;
2447 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002448
2449 ret = ops->create(dev, cd->type);
2450 if (ret < 0) {
2451 kfree(dev);
2452 return ret;
2453 }
2454
Yann Droneaud24009b02013-08-24 22:14:07 +02002455 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002456 if (ret < 0) {
2457 ops->destroy(dev);
2458 return ret;
2459 }
2460
Scott Wood07f0a7b2013-04-25 14:11:23 +00002461 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002462 kvm_get_kvm(kvm);
2463 cd->fd = ret;
2464 return 0;
2465}
2466
Alexander Graf92b591a2014-07-14 18:33:08 +02002467static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2468{
2469 switch (arg) {
2470 case KVM_CAP_USER_MEMORY:
2471 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2472 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2473#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2474 case KVM_CAP_SET_BOOT_CPU_ID:
2475#endif
2476 case KVM_CAP_INTERNAL_ERROR_DATA:
2477#ifdef CONFIG_HAVE_KVM_MSI
2478 case KVM_CAP_SIGNAL_MSI:
2479#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002480#ifdef CONFIG_HAVE_KVM_IRQFD
Paolo Bonzinidc9be0f2015-03-05 11:54:46 +01002481 case KVM_CAP_IRQFD:
Alexander Graf92b591a2014-07-14 18:33:08 +02002482 case KVM_CAP_IRQFD_RESAMPLE:
2483#endif
2484 case KVM_CAP_CHECK_EXTENSION_VM:
2485 return 1;
2486#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2487 case KVM_CAP_IRQ_ROUTING:
2488 return KVM_MAX_IRQ_ROUTES;
2489#endif
2490 default:
2491 break;
2492 }
2493 return kvm_vm_ioctl_check_extension(kvm, arg);
2494}
2495
Avi Kivitybccf2152007-02-21 18:04:26 +02002496static long kvm_vm_ioctl(struct file *filp,
2497 unsigned int ioctl, unsigned long arg)
2498{
2499 struct kvm *kvm = filp->private_data;
2500 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002501 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002502
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002503 if (kvm->mm != current->mm)
2504 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002505 switch (ioctl) {
2506 case KVM_CREATE_VCPU:
2507 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002508 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002509 case KVM_SET_USER_MEMORY_REGION: {
2510 struct kvm_userspace_memory_region kvm_userspace_mem;
2511
2512 r = -EFAULT;
2513 if (copy_from_user(&kvm_userspace_mem, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002514 sizeof(kvm_userspace_mem)))
Izik Eidus6fc138d2007-10-09 19:20:39 +02002515 goto out;
2516
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002517 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002518 break;
2519 }
2520 case KVM_GET_DIRTY_LOG: {
2521 struct kvm_dirty_log log;
2522
2523 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002524 if (copy_from_user(&log, argp, sizeof(log)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002525 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002526 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002527 break;
2528 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002529#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2530 case KVM_REGISTER_COALESCED_MMIO: {
2531 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002532
Laurent Vivier5f94c172008-05-30 16:05:54 +02002533 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002534 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02002535 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002536 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002537 break;
2538 }
2539 case KVM_UNREGISTER_COALESCED_MMIO: {
2540 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002541
Laurent Vivier5f94c172008-05-30 16:05:54 +02002542 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002543 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02002544 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002545 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002546 break;
2547 }
2548#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002549 case KVM_IRQFD: {
2550 struct kvm_irqfd data;
2551
2552 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002553 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskins721eecb2009-05-20 10:30:49 -04002554 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002555 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002556 break;
2557 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002558 case KVM_IOEVENTFD: {
2559 struct kvm_ioeventfd data;
2560
2561 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002562 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002563 goto out;
2564 r = kvm_ioeventfd(kvm, &data);
2565 break;
2566 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002567#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2568 case KVM_SET_BOOT_CPU_ID:
2569 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002570 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002571 if (atomic_read(&kvm->online_vcpus) != 0)
2572 r = -EBUSY;
2573 else
2574 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002575 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002576 break;
2577#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002578#ifdef CONFIG_HAVE_KVM_MSI
2579 case KVM_SIGNAL_MSI: {
2580 struct kvm_msi msi;
2581
2582 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002583 if (copy_from_user(&msi, argp, sizeof(msi)))
Jan Kiszka07975ad2012-03-29 21:14:12 +02002584 goto out;
2585 r = kvm_send_userspace_msi(kvm, &msi);
2586 break;
2587 }
2588#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002589#ifdef __KVM_HAVE_IRQ_LINE
2590 case KVM_IRQ_LINE_STATUS:
2591 case KVM_IRQ_LINE: {
2592 struct kvm_irq_level irq_event;
2593
2594 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002595 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002596 goto out;
2597
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002598 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2599 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002600 if (r)
2601 goto out;
2602
2603 r = -EFAULT;
2604 if (ioctl == KVM_IRQ_LINE_STATUS) {
Xiubo Li893bdbf2015-02-26 14:58:19 +08002605 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002606 goto out;
2607 }
2608
2609 r = 0;
2610 break;
2611 }
2612#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002613#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2614 case KVM_SET_GSI_ROUTING: {
2615 struct kvm_irq_routing routing;
2616 struct kvm_irq_routing __user *urouting;
2617 struct kvm_irq_routing_entry *entries;
2618
2619 r = -EFAULT;
2620 if (copy_from_user(&routing, argp, sizeof(routing)))
2621 goto out;
2622 r = -EINVAL;
2623 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2624 goto out;
2625 if (routing.flags)
2626 goto out;
2627 r = -ENOMEM;
2628 entries = vmalloc(routing.nr * sizeof(*entries));
2629 if (!entries)
2630 goto out;
2631 r = -EFAULT;
2632 urouting = argp;
2633 if (copy_from_user(entries, urouting->entries,
2634 routing.nr * sizeof(*entries)))
2635 goto out_free_irq_routing;
2636 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2637 routing.flags);
Xiubo Lia642a172015-02-26 14:58:20 +08002638out_free_irq_routing:
Alexander Grafaa8d5942013-04-15 21:12:53 +02002639 vfree(entries);
2640 break;
2641 }
2642#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002643 case KVM_CREATE_DEVICE: {
2644 struct kvm_create_device cd;
2645
2646 r = -EFAULT;
2647 if (copy_from_user(&cd, argp, sizeof(cd)))
2648 goto out;
2649
2650 r = kvm_ioctl_create_device(kvm, &cd);
2651 if (r)
2652 goto out;
2653
2654 r = -EFAULT;
2655 if (copy_to_user(argp, &cd, sizeof(cd)))
2656 goto out;
2657
2658 r = 0;
2659 break;
2660 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002661 case KVM_CHECK_EXTENSION:
2662 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2663 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002664 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002665 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002666 }
2667out:
2668 return r;
2669}
2670
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002671#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002672struct compat_kvm_dirty_log {
2673 __u32 slot;
2674 __u32 padding1;
2675 union {
2676 compat_uptr_t dirty_bitmap; /* one bit per page */
2677 __u64 padding2;
2678 };
2679};
2680
2681static long kvm_vm_compat_ioctl(struct file *filp,
2682 unsigned int ioctl, unsigned long arg)
2683{
2684 struct kvm *kvm = filp->private_data;
2685 int r;
2686
2687 if (kvm->mm != current->mm)
2688 return -EIO;
2689 switch (ioctl) {
2690 case KVM_GET_DIRTY_LOG: {
2691 struct compat_kvm_dirty_log compat_log;
2692 struct kvm_dirty_log log;
2693
2694 r = -EFAULT;
2695 if (copy_from_user(&compat_log, (void __user *)arg,
2696 sizeof(compat_log)))
2697 goto out;
2698 log.slot = compat_log.slot;
2699 log.padding1 = compat_log.padding1;
2700 log.padding2 = compat_log.padding2;
2701 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2702
2703 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002704 break;
2705 }
2706 default:
2707 r = kvm_vm_ioctl(filp, ioctl, arg);
2708 }
2709
2710out:
2711 return r;
2712}
2713#endif
2714
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002715static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002716 .release = kvm_vm_release,
2717 .unlocked_ioctl = kvm_vm_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002718#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002719 .compat_ioctl = kvm_vm_compat_ioctl,
2720#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02002721 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002722};
2723
Carsten Ottee08b9632012-01-04 10:25:20 +01002724static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002725{
Heiko Carstensaac87632010-10-27 17:22:10 +02002726 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002727 struct kvm *kvm;
2728
Carsten Ottee08b9632012-01-04 10:25:20 +01002729 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002730 if (IS_ERR(kvm))
2731 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002732#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2733 r = kvm_coalesced_mmio_init(kvm);
2734 if (r < 0) {
2735 kvm_put_kvm(kvm);
2736 return r;
2737 }
2738#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02002739 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Heiko Carstensaac87632010-10-27 17:22:10 +02002740 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002741 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002742
Heiko Carstensaac87632010-10-27 17:22:10 +02002743 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002744}
2745
2746static long kvm_dev_ioctl(struct file *filp,
2747 unsigned int ioctl, unsigned long arg)
2748{
Avi Kivity07c45a32007-03-07 13:05:38 +02002749 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002750
2751 switch (ioctl) {
2752 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002753 if (arg)
2754 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002755 r = KVM_API_VERSION;
2756 break;
2757 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002758 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002759 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002760 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02002761 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002762 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002763 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02002764 if (arg)
2765 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002766 r = PAGE_SIZE; /* struct kvm_run */
2767#ifdef CONFIG_X86
2768 r += PAGE_SIZE; /* pio data page */
2769#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002770#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2771 r += PAGE_SIZE; /* coalesced mmio ring page */
2772#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002773 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002774 case KVM_TRACE_ENABLE:
2775 case KVM_TRACE_PAUSE:
2776 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002777 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002778 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002779 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002780 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002781 }
2782out:
2783 return r;
2784}
2785
Avi Kivity6aa8b732006-12-10 02:21:36 -08002786static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002787 .unlocked_ioctl = kvm_dev_ioctl,
2788 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002789 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002790};
2791
2792static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002793 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002794 "kvm",
2795 &kvm_chardev_ops,
2796};
2797
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002798static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002799{
2800 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002801 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002802
Rusty Russell7f59f492008-12-07 21:25:45 +10302803 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002804 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002805
Rusty Russell7f59f492008-12-07 21:25:45 +10302806 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002807
Radim Krčmář13a34e02014-08-28 15:13:03 +02002808 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02002809
2810 if (r) {
2811 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2812 atomic_inc(&hardware_enable_failed);
Xiubo Li1170adc2015-02-26 14:58:26 +08002813 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
Alexander Graf10474ae2009-09-15 11:37:46 +02002814 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002815}
2816
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002817static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002818{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002819 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002820 if (kvm_usage_count)
2821 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002822 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002823}
2824
2825static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002826{
2827 int cpu = raw_smp_processor_id();
2828
Rusty Russell7f59f492008-12-07 21:25:45 +10302829 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002830 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302831 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02002832 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03002833}
2834
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002835static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002836{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002837 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002838 if (kvm_usage_count)
2839 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002840 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002841}
2842
Alexander Graf10474ae2009-09-15 11:37:46 +02002843static void hardware_disable_all_nolock(void)
2844{
2845 BUG_ON(!kvm_usage_count);
2846
2847 kvm_usage_count--;
2848 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002849 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002850}
2851
2852static void hardware_disable_all(void)
2853{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002854 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002855 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002856 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002857}
2858
2859static int hardware_enable_all(void)
2860{
2861 int r = 0;
2862
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002863 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002864
2865 kvm_usage_count++;
2866 if (kvm_usage_count == 1) {
2867 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002868 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002869
2870 if (atomic_read(&hardware_enable_failed)) {
2871 hardware_disable_all_nolock();
2872 r = -EBUSY;
2873 }
2874 }
2875
Paolo Bonzini4a937f92013-09-10 12:58:35 +02002876 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002877
2878 return r;
2879}
2880
Avi Kivity774c47f2007-02-12 00:54:47 -08002881static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2882 void *v)
2883{
2884 int cpu = (long)v;
2885
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002886 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002887 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002888 case CPU_DYING:
Xiubo Li1170adc2015-02-26 14:58:26 +08002889 pr_info("kvm: disabling virtualization on CPU%d\n",
Avi Kivity6ec8a852007-08-19 15:57:26 +03002890 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002891 hardware_disable();
Avi Kivity6ec8a852007-08-19 15:57:26 +03002892 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002893 case CPU_STARTING:
Xiubo Li1170adc2015-02-26 14:58:26 +08002894 pr_info("kvm: enabling virtualization on CPU%d\n",
Jeremy Katz43934a32007-02-19 14:37:46 +02002895 cpu);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02002896 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08002897 break;
2898 }
2899 return NOTIFY_OK;
2900}
2901
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002902static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002903 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002904{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002905 /*
2906 * Some (well, at least mine) BIOSes hang on reboot if
2907 * in vmx root mode.
2908 *
2909 * And Intel TXT required VMX off for all cpu when system shutdown.
2910 */
Xiubo Li1170adc2015-02-26 14:58:26 +08002911 pr_info("kvm: exiting hardware virtualization\n");
Sheng Yang8e1c1812009-04-29 11:09:04 +08002912 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002913 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002914 return NOTIFY_OK;
2915}
2916
2917static struct notifier_block kvm_reboot_notifier = {
2918 .notifier_call = kvm_reboot,
2919 .priority = 0,
2920};
2921
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002922static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002923{
2924 int i;
2925
2926 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002927 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002928
2929 kvm_iodevice_destructor(pos);
2930 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002931 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002932}
2933
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002934static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
Xiubo Li20e87b72015-02-26 14:58:25 +08002935 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03002936{
Sasha Levin743eeb02011-07-27 16:00:48 +03002937 if (r1->addr < r2->addr)
2938 return -1;
2939 if (r1->addr + r1->len > r2->addr + r2->len)
2940 return 1;
2941 return 0;
2942}
2943
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002944static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2945{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002946 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02002947}
2948
Geoff Levand39369f72013-04-05 19:20:30 +00002949static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03002950 gpa_t addr, int len)
2951{
Sasha Levin743eeb02011-07-27 16:00:48 +03002952 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2953 .addr = addr,
2954 .len = len,
2955 .dev = dev,
2956 };
2957
2958 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2959 kvm_io_bus_sort_cmp, NULL);
2960
2961 return 0;
2962}
2963
Geoff Levand39369f72013-04-05 19:20:30 +00002964static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03002965 gpa_t addr, int len)
2966{
2967 struct kvm_io_range *range, key;
2968 int off;
2969
2970 key = (struct kvm_io_range) {
2971 .addr = addr,
2972 .len = len,
2973 };
2974
2975 range = bsearch(&key, bus->range, bus->dev_count,
2976 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2977 if (range == NULL)
2978 return -ENOENT;
2979
2980 off = range - bus->range;
2981
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002982 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03002983 off--;
2984
2985 return off;
2986}
2987
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00002988static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
Cornelia Huck126a5af2013-07-03 16:30:53 +02002989 struct kvm_io_range *range, const void *val)
2990{
2991 int idx;
2992
2993 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2994 if (idx < 0)
2995 return -EOPNOTSUPP;
2996
2997 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02002998 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00002999 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003000 range->len, val))
3001 return idx;
3002 idx++;
3003 }
3004
3005 return -EOPNOTSUPP;
3006}
3007
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003008/* kvm_io_bus_write - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003009int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003010 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003011{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003012 struct kvm_io_bus *bus;
3013 struct kvm_io_range range;
3014 int r;
3015
3016 range = (struct kvm_io_range) {
3017 .addr = addr,
3018 .len = len,
3019 };
3020
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003021 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3022 r = __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003023 return r < 0 ? r : 0;
3024}
3025
3026/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003027int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3028 gpa_t addr, int len, const void *val, long cookie)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003029{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003030 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03003031 struct kvm_io_range range;
3032
3033 range = (struct kvm_io_range) {
3034 .addr = addr,
3035 .len = len,
3036 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003037
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003038 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003039
3040 /* First try the device referenced by cookie. */
3041 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003042 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003043 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003044 val))
3045 return cookie;
3046
3047 /*
3048 * cookie contained garbage; fall back to search and return the
3049 * correct cookie value.
3050 */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003051 return __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003052}
3053
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003054static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3055 struct kvm_io_range *range, void *val)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003056{
3057 int idx;
3058
3059 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03003060 if (idx < 0)
3061 return -EOPNOTSUPP;
3062
3063 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003064 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003065 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003066 range->len, val))
3067 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03003068 idx++;
3069 }
3070
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003071 return -EOPNOTSUPP;
3072}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03003073EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003074
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003075/* kvm_io_bus_read - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003076int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003077 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003078{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003079 struct kvm_io_bus *bus;
3080 struct kvm_io_range range;
3081 int r;
3082
3083 range = (struct kvm_io_range) {
3084 .addr = addr,
3085 .len = len,
3086 };
3087
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003088 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3089 r = __kvm_io_bus_read(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003090 return r < 0 ? r : 0;
3091}
3092
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003093
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003094/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003095int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3096 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003097{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003098 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003099
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003100 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08003101 /* exclude ioeventfd which is limited by maximum fd */
3102 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003103 return -ENOSPC;
3104
Amos Konga13007162012-03-09 12:17:32 +08003105 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3106 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003107 if (!new_bus)
3108 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003109 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3110 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003111 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003112 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3113 synchronize_srcu_expedited(&kvm->srcu);
3114 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003115
3116 return 0;
3117}
3118
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003119/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003120int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3121 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003122{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003123 int i, r;
3124 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003125
Sasha Levincdfca7b2011-12-04 19:36:28 +02003126 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003127 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003128 for (i = 0; i < bus->dev_count; i++)
3129 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003130 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003131 break;
3132 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003133
Amos Konga13007162012-03-09 12:17:32 +08003134 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003135 return r;
Amos Konga13007162012-03-09 12:17:32 +08003136
3137 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3138 sizeof(struct kvm_io_range)), GFP_KERNEL);
3139 if (!new_bus)
3140 return -ENOMEM;
3141
3142 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3143 new_bus->dev_count--;
3144 memcpy(new_bus->range + i, bus->range + i + 1,
3145 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003146
3147 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3148 synchronize_srcu_expedited(&kvm->srcu);
3149 kfree(bus);
3150 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003151}
3152
Avi Kivity774c47f2007-02-12 00:54:47 -08003153static struct notifier_block kvm_cpu_notifier = {
3154 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003155};
3156
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003157static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003158{
3159 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003160 struct kvm *kvm;
3161
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003162 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003163 spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02003164 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003165 *val += *(u32 *)((void *)kvm + offset);
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003166 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003167 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003168}
3169
3170DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3171
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003172static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003173{
3174 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003175 struct kvm *kvm;
3176 struct kvm_vcpu *vcpu;
3177 int i;
3178
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003179 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003180 spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03003181 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03003182 kvm_for_each_vcpu(i, vcpu, kvm)
3183 *val += *(u32 *)((void *)vcpu + offset);
3184
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003185 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003186 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003187}
3188
Avi Kivityba1389b2007-11-18 16:24:12 +02003189DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3190
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003191static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003192 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3193 [KVM_STAT_VM] = &vm_stat_fops,
3194};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003195
Hamo4f69b682011-12-15 14:23:16 +08003196static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003197{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003198 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003199 struct kvm_stats_debugfs_item *p;
3200
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003201 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003202 if (kvm_debugfs_dir == NULL)
3203 goto out;
3204
3205 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003206 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03003207 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02003208 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08003209 if (p->dentry == NULL)
3210 goto out_dir;
3211 }
3212
3213 return 0;
3214
3215out_dir:
3216 debugfs_remove_recursive(kvm_debugfs_dir);
3217out:
3218 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003219}
3220
3221static void kvm_exit_debug(void)
3222{
3223 struct kvm_stats_debugfs_item *p;
3224
3225 for (p = debugfs_entries; p->name; ++p)
3226 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003227 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003228}
3229
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003230static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003231{
Alexander Graf10474ae2009-09-15 11:37:46 +02003232 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003233 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003234 return 0;
3235}
3236
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003237static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003238{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003239 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003240 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003241 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003242 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003243}
3244
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003245static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003246 .suspend = kvm_suspend,
3247 .resume = kvm_resume,
3248};
3249
Avi Kivity15ad7142007-07-11 18:17:21 +03003250static inline
3251struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3252{
3253 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3254}
3255
3256static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3257{
3258 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08003259
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303260 if (vcpu->preempted)
3261 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003262
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003263 kvm_arch_sched_in(vcpu, cpu);
3264
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003265 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003266}
3267
3268static void kvm_sched_out(struct preempt_notifier *pn,
3269 struct task_struct *next)
3270{
3271 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3272
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303273 if (current->state == TASK_RUNNING)
3274 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003275 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003276}
3277
Avi Kivity0ee75be2010-04-28 15:39:01 +03003278int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003279 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003280{
3281 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003282 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003283
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003284 r = kvm_arch_init(opaque);
3285 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003286 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003287
Asias He7dac16c2013-05-08 10:57:29 +08003288 /*
3289 * kvm_arch_init makes sure there's at most one caller
3290 * for architectures that support multiple implementations,
3291 * like intel and amd on x86.
3292 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3293 * conflicts in case kvm is already setup for another implementation.
3294 */
3295 r = kvm_irqfd_init();
3296 if (r)
3297 goto out_irqfd;
3298
Avi Kivity8437a612009-06-06 14:52:35 -07003299 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303300 r = -ENOMEM;
3301 goto out_free_0;
3302 }
3303
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003304 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003305 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303306 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003307
Yang, Sheng002c7f72007-07-31 14:23:01 +03003308 for_each_online_cpu(cpu) {
3309 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003310 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003311 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003312 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003313 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003314 }
3315
Avi Kivity774c47f2007-02-12 00:54:47 -08003316 r = register_cpu_notifier(&kvm_cpu_notifier);
3317 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003318 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003319 register_reboot_notifier(&kvm_reboot_notifier);
3320
Rusty Russellc16f8622007-07-30 21:12:19 +10003321 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003322 if (!vcpu_align)
3323 vcpu_align = __alignof__(struct kvm_vcpu);
3324 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003325 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003326 if (!kvm_vcpu_cache) {
3327 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003328 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003329 }
3330
Gleb Natapovaf585b92010-10-14 11:22:46 +02003331 r = kvm_async_pf_init();
3332 if (r)
3333 goto out_free;
3334
Avi Kivity6aa8b732006-12-10 02:21:36 -08003335 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003336 kvm_vm_fops.owner = module;
3337 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003338
3339 r = misc_register(&kvm_dev);
3340 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003341 pr_err("kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003342 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003343 }
3344
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003345 register_syscore_ops(&kvm_syscore_ops);
3346
Avi Kivity15ad7142007-07-11 18:17:21 +03003347 kvm_preempt_ops.sched_in = kvm_sched_in;
3348 kvm_preempt_ops.sched_out = kvm_sched_out;
3349
Hamo4f69b682011-12-15 14:23:16 +08003350 r = kvm_init_debug();
3351 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003352 pr_err("kvm: create debugfs files failed\n");
Hamo4f69b682011-12-15 14:23:16 +08003353 goto out_undebugfs;
3354 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003355
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003356 r = kvm_vfio_ops_init();
3357 WARN_ON(r);
3358
Avi Kivityc7addb92007-09-16 18:58:32 +02003359 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003360
Hamo4f69b682011-12-15 14:23:16 +08003361out_undebugfs:
3362 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003363 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003364out_unreg:
3365 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003366out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003367 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003368out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003369 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003370 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003371out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003372out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003373 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303374out_free_0a:
3375 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003376out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003377 kvm_irqfd_exit();
3378out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003379 kvm_arch_exit();
3380out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003381 return r;
3382}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003383EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003384
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003385void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003386{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003387 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003388 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003389 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003390 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003391 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003392 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003393 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003394 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003395 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003396 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003397 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303398 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08003399 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003400}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003401EXPORT_SYMBOL_GPL(kvm_exit);