blob: be70035fd42a14edad4c24ae411c4b9d3fdf396c [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Jan Kiszkae935b832011-02-08 12:55:33 +010073DEFINE_RAW_SPINLOCK(kvm_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080075
Rusty Russell7f59f492008-12-07 21:25:45 +103076static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020077static int kvm_usage_count = 0;
78static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030079
Rusty Russellc16f8622007-07-30 21:12:19 +100080struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030082
Avi Kivity15ad7142007-07-11 18:17:21 +030083static __read_mostly struct preempt_ops kvm_preempt_ops;
84
Hollis Blanchard76f7c872008-04-15 16:05:42 -050085struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080086
Avi Kivitybccf2152007-02-21 18:04:26 +020087static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020089#ifdef CONFIG_COMPAT
90static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020093static int hardware_enable_all(void);
94static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020095
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020096static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
Avi Kivityb7c41452010-12-02 17:52:50 +020098bool kvm_rebooting;
99EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300100
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300101static bool largepages_enabled = true;
102
Xiao Guangronga2766322012-07-26 11:58:59 +0800103bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300104{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100105 if (pfn_valid(pfn)) {
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800106 int reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800107 struct page *tail = pfn_to_page(pfn);
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800110 if (head != tail) {
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800111 /*
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
119 * what we just read.
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800120 */
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800121 smp_rmb();
122 if (PageTail(tail))
123 return reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800124 }
125 return PageReserved(tail);
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100126 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300127
128 return true;
129}
130
Avi Kivity6aa8b732006-12-10 02:21:36 -0800131/*
132 * Switches to specified vcpu, until a matching vcpu_put()
133 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300134int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135{
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 int cpu;
137
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300138 if (mutex_lock_killable(&vcpu->mutex))
139 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
145 synchronize_rcu();
146 put_pid(oldpid);
147 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300148 cpu = get_cpu();
149 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200150 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300151 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300152 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200153}
154
Carsten Otte313a3dc2007-10-11 19:16:52 +0200155void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800156{
Avi Kivity15ad7142007-07-11 18:17:21 +0300157 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200158 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300159 preempt_notifier_unregister(&vcpu->preempt_notifier);
160 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800161 mutex_unlock(&vcpu->mutex);
162}
163
Avi Kivityd9e368d2007-06-07 19:18:30 +0300164static void ack_flush(void *_completed)
165{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300166}
167
Rusty Russell49846892008-12-08 20:26:24 +1030168static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300169{
Avi Kivity597a5f52008-07-20 14:24:22 +0300170 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030171 cpumask_var_t cpus;
172 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300173 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300174
Li Zefan79f55992009-06-15 14:58:26 +0800175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800177 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300178 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800179 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300180 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800181
182 /* Set ->requests bit before we read ->mode */
183 smp_mb();
184
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030187 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300188 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
193 else
194 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800195 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030196 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030197 return called;
198}
199
200void kvm_flush_remote_tlbs(struct kvm *kvm)
201{
Alex Shibec87d62012-03-04 13:13:38 +0800202 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800203
204 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300208}
209
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500210void kvm_reload_remote_mmus(struct kvm *kvm)
211{
Rusty Russell49846892008-12-08 20:26:24 +1030212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500213}
214
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000215int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
216{
217 struct page *page;
218 int r;
219
220 mutex_init(&vcpu->mutex);
221 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000222 vcpu->kvm = kvm;
223 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500224 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300225 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200226 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000227
228 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
229 if (!page) {
230 r = -ENOMEM;
231 goto fail;
232 }
233 vcpu->run = page_address(page);
234
Raghavendra K T4c088492012-07-18 19:07:46 +0530235 kvm_vcpu_set_in_spin_loop(vcpu, false);
236 kvm_vcpu_set_dy_eligible(vcpu, false);
237
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800238 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000239 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800240 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000241 return 0;
242
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000243fail_free_run:
244 free_page((unsigned long)vcpu->run);
245fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000246 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000247}
248EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500252 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800253 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000254 free_page((unsigned long)vcpu->run);
255}
256EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200258#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260{
261 return container_of(mn, struct kvm, mmu_notifier);
262}
263
264static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
267{
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200269 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200270
271 /*
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * before returning.
280 *
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
283 *
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
288 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200289 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200290 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900291
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200292 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200294 /* we've to flush the tlb before the pages can be freed */
295 if (need_tlb_flush)
296 kvm_flush_remote_tlbs(kvm);
297
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900298 spin_unlock(&kvm->mmu_lock);
299 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200300}
301
Izik Eidus3da0dd42009-09-23 21:47:18 +0300302static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
305 pte_t pte)
306{
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200308 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300309
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200310 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200315 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300316}
317
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200318static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
320 unsigned long start,
321 unsigned long end)
322{
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200324 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200325
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200326 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200327 spin_lock(&kvm->mmu_lock);
328 /*
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
332 */
333 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800335 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200336 /* we've to flush the tlb before the pages can be freed */
337 if (need_tlb_flush)
338 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900339
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200342}
343
344static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
346 unsigned long start,
347 unsigned long end)
348{
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350
351 spin_lock(&kvm->mmu_lock);
352 /*
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
355 * been freed.
356 */
357 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000358 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200359 /*
360 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200363 */
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
366
367 BUG_ON(kvm->mmu_notifier_count < 0);
368}
369
370static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long address)
373{
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
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900380 young = kvm_age_hva(kvm, address);
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)
461 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200462
463 r = hardware_enable_all();
464 if (r)
465 goto out_err_nodisable;
466
Avi Kivity75858a82009-01-04 17:10:50 +0200467#ifdef CONFIG_HAVE_KVM_IRQCHIP
468 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300469 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200470#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800471
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200472 r = -ENOMEM;
473 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
474 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100475 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800476 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200477 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100478 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200479 for (i = 0; i < KVM_NR_BUSES; i++) {
480 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
481 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100482 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200483 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200484 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700485
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900486 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200487 kvm->mm = current->mm;
488 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400489 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800490 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300491 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200492 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300493 atomic_set(&kvm->users_count, 1);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700494
495 r = kvm_init_mmu_notifier(kvm);
496 if (r)
497 goto out_err;
498
Jan Kiszkae935b832011-02-08 12:55:33 +0100499 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000500 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100501 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100502
Avi Kivityf17abe92007-02-21 19:28:04 +0200503 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200504
505out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100506 cleanup_srcu_struct(&kvm->srcu);
507out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200508 hardware_disable_all();
509out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200510 for (i = 0; i < KVM_NR_BUSES; i++)
511 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200512 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100513 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200514 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200515}
516
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900517/*
518 * Avoid using vmalloc for a small buffer.
519 * Should not be used when the size is statically known.
520 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900521void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900522{
523 if (size > PAGE_SIZE)
524 return vzalloc(size);
525 else
526 return kzalloc(size, GFP_KERNEL);
527}
528
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900529void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900530{
531 if (is_vmalloc_addr(addr))
532 vfree(addr);
533 else
534 kfree(addr);
535}
536
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900537static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538{
539 if (!memslot->dirty_bitmap)
540 return;
541
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900542 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900543 memslot->dirty_bitmap = NULL;
544}
545
Avi Kivity6aa8b732006-12-10 02:21:36 -0800546/*
547 * Free any memory in @free but not in @dont.
548 */
549static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
550 struct kvm_memory_slot *dont)
551{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800552 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900553 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800554
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900555 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300556
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800558}
559
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800560void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800561{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200562 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800563 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800564
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800565 kvm_for_each_memslot(memslot, slots)
566 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200567
568 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569}
570
Avi Kivityf17abe92007-02-21 19:28:04 +0200571static void kvm_destroy_vm(struct kvm *kvm)
572{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200573 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200574 struct mm_struct *mm = kvm->mm;
575
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800576 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100577 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800578 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100579 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200580 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200581 for (i = 0; i < KVM_NR_BUSES; i++)
582 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200583 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200584#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
585 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200586#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300587 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200588#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800589 kvm_arch_destroy_vm(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100590 kvm_free_physmem(kvm);
591 cleanup_srcu_struct(&kvm->srcu);
592 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200593 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200594 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200595}
596
Izik Eidusd39f13b2008-03-30 16:01:25 +0300597void kvm_get_kvm(struct kvm *kvm)
598{
599 atomic_inc(&kvm->users_count);
600}
601EXPORT_SYMBOL_GPL(kvm_get_kvm);
602
603void kvm_put_kvm(struct kvm *kvm)
604{
605 if (atomic_dec_and_test(&kvm->users_count))
606 kvm_destroy_vm(kvm);
607}
608EXPORT_SYMBOL_GPL(kvm_put_kvm);
609
610
Avi Kivityf17abe92007-02-21 19:28:04 +0200611static int kvm_vm_release(struct inode *inode, struct file *filp)
612{
613 struct kvm *kvm = filp->private_data;
614
Gregory Haskins721eecb2009-05-20 10:30:49 -0400615 kvm_irqfd_release(kvm);
616
Izik Eidusd39f13b2008-03-30 16:01:25 +0300617 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800618 return 0;
619}
620
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900621/*
622 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900623 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900624 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900625static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900627#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900628 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900629
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900630 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900631 if (!memslot->dirty_bitmap)
632 return -ENOMEM;
633
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900634#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900635 return 0;
636}
637
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800638static int cmp_memslot(const void *slot1, const void *slot2)
639{
640 struct kvm_memory_slot *s1, *s2;
641
642 s1 = (struct kvm_memory_slot *)slot1;
643 s2 = (struct kvm_memory_slot *)slot2;
644
645 if (s1->npages < s2->npages)
646 return 1;
647 if (s1->npages > s2->npages)
648 return -1;
649
650 return 0;
651}
652
653/*
654 * Sort the memslots base on its size, so the larger slots
655 * will get better fit.
656 */
657static void sort_memslots(struct kvm_memslots *slots)
658{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800659 int i;
660
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800661 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
662 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800663
664 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
665 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800666}
667
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800668void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
669{
670 if (new) {
671 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800672 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800673 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800674
Xiao Guangrong28a37542011-11-24 19:04:35 +0800675 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800676 if (new->npages != npages)
677 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800678 }
679
680 slots->generation++;
681}
682
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800683static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
684{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800685 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
686
687#ifdef KVM_CAP_READONLY_MEM
688 valid_flags |= KVM_MEM_READONLY;
689#endif
690
691 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800692 return -EINVAL;
693
694 return 0;
695}
696
Avi Kivity6aa8b732006-12-10 02:21:36 -0800697/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800698 * Allocate some memory and give it an address in the guest physical address
699 * space.
700 *
701 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800702 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500703 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800704 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800705int __kvm_set_memory_region(struct kvm *kvm,
706 struct kvm_userspace_memory_region *mem,
707 int user_alloc)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800708{
Avi Kivity8234b222010-12-27 12:08:45 +0200709 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800710 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200711 unsigned long npages;
712 unsigned long i;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800713 struct kvm_memory_slot *memslot;
714 struct kvm_memory_slot old, new;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200715 struct kvm_memslots *slots, *old_memslots;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800716
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800717 r = check_memory_region_flags(mem);
718 if (r)
719 goto out;
720
Avi Kivity6aa8b732006-12-10 02:21:36 -0800721 r = -EINVAL;
722 /* General sanity checks */
723 if (mem->memory_size & (PAGE_SIZE - 1))
724 goto out;
725 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
726 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900727 /* We can read the guest memory with __xxx_user() later on. */
728 if (user_alloc &&
729 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200730 !access_ok(VERIFY_WRITE,
731 (void __user *)(unsigned long)mem->userspace_addr,
732 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600733 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800734 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800735 goto out;
736 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
737 goto out;
738
Xiao Guangrong28a37542011-11-24 19:04:35 +0800739 memslot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800740 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741 npages = mem->memory_size >> PAGE_SHIFT;
742
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900743 r = -EINVAL;
744 if (npages > KVM_MEM_MAX_NR_PAGES)
745 goto out;
746
Avi Kivity6aa8b732006-12-10 02:21:36 -0800747 if (!npages)
748 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
749
Avi Kivity6aa8b732006-12-10 02:21:36 -0800750 new = old = *memslot;
751
Avi Kivitye36d96f2010-06-21 10:56:36 +0300752 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800753 new.base_gfn = base_gfn;
754 new.npages = npages;
755 new.flags = mem->flags;
756
757 /* Disallow changing a memory slot's size. */
758 r = -EINVAL;
759 if (npages && old.npages && npages != old.npages)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800760 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800761
762 /* Check for overlaps */
763 r = -EEXIST;
764 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200765 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
Avi Kivity6aa8b732006-12-10 02:21:36 -0800766
Jan Kiszka4cd481f2009-04-13 11:59:32 +0200767 if (s == memslot || !s->npages)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800768 continue;
769 if (!((base_gfn + npages <= s->base_gfn) ||
770 (base_gfn >= s->base_gfn + s->npages)))
Sheng Yangf78e0e22007-10-29 09:40:42 +0800771 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800772 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800773
Avi Kivity6aa8b732006-12-10 02:21:36 -0800774 /* Free page dirty bitmap if unneeded */
775 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000776 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800777
778 r = -ENOMEM;
779
780 /* Allocate if a slot is being created */
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900781 if (npages && !old.npages) {
782 new.user_alloc = user_alloc;
783 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900784
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900785 if (kvm_arch_create_memslot(&new, npages))
786 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800787 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200788
Avi Kivity6aa8b732006-12-10 02:21:36 -0800789 /* Allocate page dirty bitmap if needed */
790 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900791 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800792 goto out_free;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200793 /* destroy any largepage mappings for dirty tracking */
Avi Kivity6aa8b732006-12-10 02:21:36 -0800794 }
795
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300796 if (!npages || base_gfn != old.base_gfn) {
Xiao Guangrong28a37542011-11-24 19:04:35 +0800797 struct kvm_memory_slot *slot;
798
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200799 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100800 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
801 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200802 if (!slots)
803 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800804 slot = id_to_memslot(slots, mem->slot);
805 slot->flags |= KVM_MEMSLOT_INVALID;
806
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800807 update_memslots(slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200808
809 old_memslots = kvm->memslots;
810 rcu_assign_pointer(kvm->memslots, slots);
811 synchronize_srcu_expedited(&kvm->srcu);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300812 /* From this point no new shadow pages pointing to a deleted,
813 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200814 *
815 * validation of sp->gfn happens in:
816 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
817 * - kvm_is_visible_gfn (mmu_check_roots)
818 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300819 kvm_arch_flush_shadow_memslot(kvm, slot);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200820 kfree(old_memslots);
821 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300822
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200823 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
824 if (r)
825 goto out_free;
826
Alex Williamson32f6daa2012-04-11 09:51:49 -0600827 /* map/unmap the pages in iommu page table */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200828 if (npages) {
829 r = kvm_iommu_map_pages(kvm, &new);
830 if (r)
831 goto out_free;
Alex Williamson32f6daa2012-04-11 09:51:49 -0600832 } else
833 kvm_iommu_unmap_pages(kvm, &old);
Andrea Arcangeli604b38a2008-07-25 16:32:03 +0200834
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200835 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100836 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
837 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200838 if (!slots)
839 goto out_free;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200840
841 /* actual memory is freed via old in kvm_free_physmem_slot below */
842 if (!npages) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200843 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900844 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200845 }
846
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800847 update_memslots(slots, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200848 old_memslots = kvm->memslots;
849 rcu_assign_pointer(kvm->memslots, slots);
850 synchronize_srcu_expedited(&kvm->srcu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800851
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200852 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800853
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200854 kvm_free_physmem_slot(&old, &new);
855 kfree(old_memslots);
856
Avi Kivity6aa8b732006-12-10 02:21:36 -0800857 return 0;
858
Sheng Yangf78e0e22007-10-29 09:40:42 +0800859out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800860 kvm_free_physmem_slot(&new, &old);
861out:
862 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200863
864}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800865EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
866
867int kvm_set_memory_region(struct kvm *kvm,
868 struct kvm_userspace_memory_region *mem,
869 int user_alloc)
870{
871 int r;
872
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200873 mutex_lock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800874 r = __kvm_set_memory_region(kvm, mem, user_alloc);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200875 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800876 return r;
877}
Izik Eidus210c7c42007-10-24 23:52:57 +0200878EXPORT_SYMBOL_GPL(kvm_set_memory_region);
879
Carsten Otte1fe779f2007-10-29 16:08:35 +0100880int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
881 struct
882 kvm_userspace_memory_region *mem,
883 int user_alloc)
Izik Eidus210c7c42007-10-24 23:52:57 +0200884{
Izik Eiduse0d62c72007-10-24 23:57:46 +0200885 if (mem->slot >= KVM_MEMORY_SLOTS)
886 return -EINVAL;
Izik Eidus210c7c42007-10-24 23:52:57 +0200887 return kvm_set_memory_region(kvm, mem, user_alloc);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800888}
889
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800890int kvm_get_dirty_log(struct kvm *kvm,
891 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800892{
893 struct kvm_memory_slot *memslot;
894 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900895 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800896 unsigned long any = 0;
897
Avi Kivity6aa8b732006-12-10 02:21:36 -0800898 r = -EINVAL;
899 if (log->slot >= KVM_MEMORY_SLOTS)
900 goto out;
901
Xiao Guangrong28a37542011-11-24 19:04:35 +0800902 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800903 r = -ENOENT;
904 if (!memslot->dirty_bitmap)
905 goto out;
906
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900907 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800908
Uri Lublincd1a4a92007-02-22 16:43:09 +0200909 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800910 any = memslot->dirty_bitmap[i];
911
912 r = -EFAULT;
913 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
914 goto out;
915
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800916 if (any)
917 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800918
919 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800920out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800921 return r;
922}
923
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900924bool kvm_largepages_enabled(void)
925{
926 return largepages_enabled;
927}
928
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300929void kvm_disable_largepages(void)
930{
931 largepages_enabled = false;
932}
933EXPORT_SYMBOL_GPL(kvm_disable_largepages);
934
Gleb Natapov49c77542010-10-18 15:22:23 +0200935struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
936{
937 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
938}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300939EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800940
Izik Eiduse0d62c72007-10-24 23:57:46 +0200941int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
942{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800943 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +0200944
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800945 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
946 memslot->flags & KVM_MEMSLOT_INVALID)
947 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200948
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800949 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200950}
951EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
952
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100953unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
954{
955 struct vm_area_struct *vma;
956 unsigned long addr, size;
957
958 size = PAGE_SIZE;
959
960 addr = gfn_to_hva(kvm, gfn);
961 if (kvm_is_error_hva(addr))
962 return PAGE_SIZE;
963
964 down_read(&current->mm->mmap_sem);
965 vma = find_vma(current->mm, addr);
966 if (!vma)
967 goto out;
968
969 size = vma_kernel_pagesize(vma);
970
971out:
972 up_read(&current->mm->mmap_sem);
973
974 return size;
975}
976
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800977static bool memslot_is_readonly(struct kvm_memory_slot *slot)
978{
979 return slot->flags & KVM_MEM_READONLY;
980}
981
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800982static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
983 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +0200984{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200985 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +0800986 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +0800987
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800988 if (memslot_is_readonly(slot) && write)
989 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +0800990
991 if (nr_pages)
992 *nr_pages = slot->npages - (gfn - slot->base_gfn);
993
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800994 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +0200995}
Xiao Guangrong48987782010-08-22 19:11:43 +0800996
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800997static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
998 gfn_t *nr_pages)
999{
1000 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1001}
1002
1003unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1004 gfn_t gfn)
1005{
1006 return gfn_to_hva_many(slot, gfn, NULL);
1007}
1008EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1009
Xiao Guangrong48987782010-08-22 19:11:43 +08001010unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1011{
Gleb Natapov49c77542010-10-18 15:22:23 +02001012 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001013}
Sheng Yang0d150292008-04-25 21:44:50 +08001014EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001015
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001016/*
1017 * The hva returned by this function is only allowed to be read.
1018 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1019 */
1020static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001021{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001022 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001023}
1024
1025static int kvm_read_hva(void *data, void __user *hva, int len)
1026{
1027 return __copy_from_user(data, hva, len);
1028}
1029
1030static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1031{
1032 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001033}
1034
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001035int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1036 unsigned long start, int write, struct page **page)
1037{
1038 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1039
1040 if (write)
1041 flags |= FOLL_WRITE;
1042
1043 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1044}
1045
Huang Yingfafc3db2011-01-30 11:15:49 +08001046static inline int check_user_page_hwpoison(unsigned long addr)
1047{
1048 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1049
1050 rc = __get_user_pages(current, current->mm, addr, 1,
1051 flags, NULL, NULL, NULL);
1052 return rc == -EHWPOISON;
1053}
1054
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001055/*
1056 * The atomic path to get the writable pfn which will be stored in @pfn,
1057 * true indicates success, otherwise false is returned.
1058 */
1059static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1060 bool write_fault, bool *writable, pfn_t *pfn)
1061{
1062 struct page *page[1];
1063 int npages;
1064
1065 if (!(async || atomic))
1066 return false;
1067
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001068 /*
1069 * Fast pin a writable pfn only if it is a write fault request
1070 * or the caller allows to map a writable pfn for a read fault
1071 * request.
1072 */
1073 if (!(write_fault || writable))
1074 return false;
1075
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001076 npages = __get_user_pages_fast(addr, 1, 1, page);
1077 if (npages == 1) {
1078 *pfn = page_to_pfn(page[0]);
1079
1080 if (writable)
1081 *writable = true;
1082 return true;
1083 }
1084
1085 return false;
1086}
1087
1088/*
1089 * The slow path to get the pfn of the specified host virtual address,
1090 * 1 indicates success, -errno is returned if error is detected.
1091 */
1092static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1093 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001094{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001095 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001096 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001097
1098 might_sleep();
1099
1100 if (writable)
1101 *writable = write_fault;
1102
1103 if (async) {
1104 down_read(&current->mm->mmap_sem);
1105 npages = get_user_page_nowait(current, current->mm,
1106 addr, write_fault, page);
1107 up_read(&current->mm->mmap_sem);
1108 } else
1109 npages = get_user_pages_fast(addr, 1, write_fault,
1110 page);
1111 if (npages != 1)
1112 return npages;
1113
1114 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001115 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001116 struct page *wpage[1];
1117
1118 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119 if (npages == 1) {
1120 *writable = true;
1121 put_page(page[0]);
1122 page[0] = wpage[0];
1123 }
1124
1125 npages = 1;
1126 }
1127 *pfn = page_to_pfn(page[0]);
1128 return npages;
1129}
1130
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001131static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1132{
1133 if (unlikely(!(vma->vm_flags & VM_READ)))
1134 return false;
1135
1136 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1137 return false;
1138
1139 return true;
1140}
1141
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001142/*
1143 * Pin guest page in memory and return its pfn.
1144 * @addr: host virtual address which maps memory to the guest
1145 * @atomic: whether this function can sleep
1146 * @async: whether this function need to wait IO complete if the
1147 * host page is not in the memory
1148 * @write_fault: whether we should get a writable host page
1149 * @writable: whether it allows to map a writable host page for !@write_fault
1150 *
1151 * The function will map a writable host page for these two cases:
1152 * 1): @write_fault = true
1153 * 2): @write_fault = false && @writable, @writable will tell the caller
1154 * whether the mapping is writable.
1155 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001156static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1157 bool write_fault, bool *writable)
1158{
1159 struct vm_area_struct *vma;
1160 pfn_t pfn = 0;
1161 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001162
Gleb Natapovaf585b92010-10-14 11:22:46 +02001163 /* we can do it either atomically or asynchronously, not both */
1164 BUG_ON(atomic && async);
1165
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001166 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1167 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001168
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001169 if (atomic)
1170 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001171
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001172 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1173 if (npages == 1)
1174 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001175
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001176 down_read(&current->mm->mmap_sem);
1177 if (npages == -EHWPOISON ||
1178 (!async && check_user_page_hwpoison(addr))) {
1179 pfn = KVM_PFN_ERR_HWPOISON;
1180 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001181 }
Izik Eidus539cb662007-11-11 22:05:04 +02001182
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001183 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001184
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001185 if (vma == NULL)
1186 pfn = KVM_PFN_ERR_FAULT;
1187 else if ((vma->vm_flags & VM_PFNMAP)) {
1188 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1189 vma->vm_pgoff;
1190 BUG_ON(!kvm_is_mmio_pfn(pfn));
1191 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001192 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001193 *async = true;
1194 pfn = KVM_PFN_ERR_FAULT;
1195 }
1196exit:
1197 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001198 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001199}
1200
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001201static pfn_t
1202__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1203 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001204{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001205 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1206
1207 if (addr == KVM_HVA_ERR_RO_BAD)
1208 return KVM_PFN_ERR_RO_FAULT;
1209
1210 if (kvm_is_error_hva(addr))
1211 return KVM_PFN_ERR_BAD;
1212
1213 /* Do not map writable pfn in the readonly memslot. */
1214 if (writable && memslot_is_readonly(slot)) {
1215 *writable = false;
1216 writable = NULL;
1217 }
1218
1219 return hva_to_pfn(addr, atomic, async, write_fault,
1220 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001221}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001222
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001223static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1224 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001225{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001226 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001227
Gleb Natapovaf585b92010-10-14 11:22:46 +02001228 if (async)
1229 *async = false;
1230
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001231 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001232
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001233 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1234 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001235}
1236
1237pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1238{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001239 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001240}
1241EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1242
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001243pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1244 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001245{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001246 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001247}
1248EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1249
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001250pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1251{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001252 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001253}
Anthony Liguori35149e22008-04-02 14:46:56 -05001254EXPORT_SYMBOL_GPL(gfn_to_pfn);
1255
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001256pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1257 bool *writable)
1258{
1259 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1260}
1261EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1262
Xiao Guangrongd5661042012-07-17 21:56:16 +08001263pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001264{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001265 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001266}
1267
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001268pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1269{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001270 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001271}
1272EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1273
Xiao Guangrong48987782010-08-22 19:11:43 +08001274int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1275 int nr_pages)
1276{
1277 unsigned long addr;
1278 gfn_t entry;
1279
Gleb Natapov49c77542010-10-18 15:22:23 +02001280 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001281 if (kvm_is_error_hva(addr))
1282 return -1;
1283
1284 if (entry < nr_pages)
1285 return 0;
1286
1287 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1288}
1289EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1290
Xiao Guangronga2766322012-07-26 11:58:59 +08001291static struct page *kvm_pfn_to_page(pfn_t pfn)
1292{
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001293 if (is_error_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001294 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001295
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001296 if (kvm_is_mmio_pfn(pfn)) {
1297 WARN_ON(1);
1298 return KVM_ERR_PTR_BAD_PAGE;
1299 }
1300
Xiao Guangronga2766322012-07-26 11:58:59 +08001301 return pfn_to_page(pfn);
1302}
1303
Anthony Liguori35149e22008-04-02 14:46:56 -05001304struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1305{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001306 pfn_t pfn;
1307
1308 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001309
Xiao Guangronga2766322012-07-26 11:58:59 +08001310 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001311}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001312
Avi Kivity954bbbc2007-03-30 14:02:32 +03001313EXPORT_SYMBOL_GPL(gfn_to_page);
1314
Izik Eidusb4231d62007-11-20 11:49:33 +02001315void kvm_release_page_clean(struct page *page)
1316{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001317 WARN_ON(is_error_page(page));
1318
Anthony Liguori35149e22008-04-02 14:46:56 -05001319 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001320}
1321EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1322
Anthony Liguori35149e22008-04-02 14:46:56 -05001323void kvm_release_pfn_clean(pfn_t pfn)
1324{
Xiao Guangrongf3ac1a42012-10-16 20:07:03 +08001325 if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001326 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001327}
1328EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1329
Izik Eidusb4231d62007-11-20 11:49:33 +02001330void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001331{
Xiao Guangronga2766322012-07-26 11:58:59 +08001332 WARN_ON(is_error_page(page));
1333
Anthony Liguori35149e22008-04-02 14:46:56 -05001334 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001335}
Izik Eidusb4231d62007-11-20 11:49:33 +02001336EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001337
Anthony Liguori35149e22008-04-02 14:46:56 -05001338void kvm_release_pfn_dirty(pfn_t pfn)
1339{
1340 kvm_set_pfn_dirty(pfn);
1341 kvm_release_pfn_clean(pfn);
1342}
1343EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1344
1345void kvm_set_page_dirty(struct page *page)
1346{
1347 kvm_set_pfn_dirty(page_to_pfn(page));
1348}
1349EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1350
1351void kvm_set_pfn_dirty(pfn_t pfn)
1352{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001353 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001354 struct page *page = pfn_to_page(pfn);
1355 if (!PageReserved(page))
1356 SetPageDirty(page);
1357 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001358}
1359EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1360
1361void kvm_set_pfn_accessed(pfn_t pfn)
1362{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001363 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001364 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001365}
1366EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1367
1368void kvm_get_pfn(pfn_t pfn)
1369{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001370 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001371 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001372}
1373EXPORT_SYMBOL_GPL(kvm_get_pfn);
1374
Izik Eidus195aefd2007-10-01 22:14:18 +02001375static int next_segment(unsigned long len, int offset)
1376{
1377 if (len > PAGE_SIZE - offset)
1378 return PAGE_SIZE - offset;
1379 else
1380 return len;
1381}
1382
1383int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1384 int len)
1385{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001386 int r;
1387 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001388
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001389 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001390 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001391 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001392 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001393 if (r)
1394 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001395 return 0;
1396}
1397EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1398
1399int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1400{
1401 gfn_t gfn = gpa >> PAGE_SHIFT;
1402 int seg;
1403 int offset = offset_in_page(gpa);
1404 int ret;
1405
1406 while ((seg = next_segment(len, offset)) != 0) {
1407 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1408 if (ret < 0)
1409 return ret;
1410 offset = 0;
1411 len -= seg;
1412 data += seg;
1413 ++gfn;
1414 }
1415 return 0;
1416}
1417EXPORT_SYMBOL_GPL(kvm_read_guest);
1418
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001419int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1420 unsigned long len)
1421{
1422 int r;
1423 unsigned long addr;
1424 gfn_t gfn = gpa >> PAGE_SHIFT;
1425 int offset = offset_in_page(gpa);
1426
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001427 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001428 if (kvm_is_error_hva(addr))
1429 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001430 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001431 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001432 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001433 if (r)
1434 return -EFAULT;
1435 return 0;
1436}
1437EXPORT_SYMBOL(kvm_read_guest_atomic);
1438
Izik Eidus195aefd2007-10-01 22:14:18 +02001439int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1440 int offset, int len)
1441{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001442 int r;
1443 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001444
Izik Eiduse0506bc2007-11-11 22:10:22 +02001445 addr = gfn_to_hva(kvm, gfn);
1446 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001447 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001448 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001449 if (r)
1450 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001451 mark_page_dirty(kvm, gfn);
1452 return 0;
1453}
1454EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1455
1456int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1457 unsigned long len)
1458{
1459 gfn_t gfn = gpa >> PAGE_SHIFT;
1460 int seg;
1461 int offset = offset_in_page(gpa);
1462 int ret;
1463
1464 while ((seg = next_segment(len, offset)) != 0) {
1465 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1466 if (ret < 0)
1467 return ret;
1468 offset = 0;
1469 len -= seg;
1470 data += seg;
1471 ++gfn;
1472 }
1473 return 0;
1474}
1475
Gleb Natapov49c77542010-10-18 15:22:23 +02001476int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1477 gpa_t gpa)
1478{
1479 struct kvm_memslots *slots = kvm_memslots(kvm);
1480 int offset = offset_in_page(gpa);
1481 gfn_t gfn = gpa >> PAGE_SHIFT;
1482
1483 ghc->gpa = gpa;
1484 ghc->generation = slots->generation;
Paul Mackerras9d4cba72012-01-12 20:09:51 +00001485 ghc->memslot = gfn_to_memslot(kvm, gfn);
Gleb Natapov49c77542010-10-18 15:22:23 +02001486 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1487 if (!kvm_is_error_hva(ghc->hva))
1488 ghc->hva += offset;
1489 else
1490 return -EFAULT;
1491
1492 return 0;
1493}
1494EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1495
1496int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1497 void *data, unsigned long len)
1498{
1499 struct kvm_memslots *slots = kvm_memslots(kvm);
1500 int r;
1501
1502 if (slots->generation != ghc->generation)
1503 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1504
1505 if (kvm_is_error_hva(ghc->hva))
1506 return -EFAULT;
1507
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001508 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001509 if (r)
1510 return -EFAULT;
1511 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1512
1513 return 0;
1514}
1515EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1516
Gleb Natapove03b6442011-07-11 15:28:11 -04001517int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1518 void *data, unsigned long len)
1519{
1520 struct kvm_memslots *slots = kvm_memslots(kvm);
1521 int r;
1522
1523 if (slots->generation != ghc->generation)
1524 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1525
1526 if (kvm_is_error_hva(ghc->hva))
1527 return -EFAULT;
1528
1529 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1530 if (r)
1531 return -EFAULT;
1532
1533 return 0;
1534}
1535EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1536
Izik Eidus195aefd2007-10-01 22:14:18 +02001537int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1538{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001539 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1540 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001541}
1542EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1543
1544int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1545{
1546 gfn_t gfn = gpa >> PAGE_SHIFT;
1547 int seg;
1548 int offset = offset_in_page(gpa);
1549 int ret;
1550
1551 while ((seg = next_segment(len, offset)) != 0) {
1552 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1553 if (ret < 0)
1554 return ret;
1555 offset = 0;
1556 len -= seg;
1557 ++gfn;
1558 }
1559 return 0;
1560}
1561EXPORT_SYMBOL_GPL(kvm_clear_guest);
1562
Gleb Natapov49c77542010-10-18 15:22:23 +02001563void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1564 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001565{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001566 if (memslot && memslot->dirty_bitmap) {
1567 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001568
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001569 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001570 }
1571}
1572
Gleb Natapov49c77542010-10-18 15:22:23 +02001573void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1574{
1575 struct kvm_memory_slot *memslot;
1576
1577 memslot = gfn_to_memslot(kvm, gfn);
1578 mark_page_dirty_in_slot(kvm, memslot, gfn);
1579}
1580
Eddie Dongb6958ce2007-07-18 12:15:21 +03001581/*
1582 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1583 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001584void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001585{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001586 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001587
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001588 for (;;) {
1589 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001590
Gleb Natapova1b37102009-07-09 15:33:52 +03001591 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001592 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001593 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001594 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001595 if (kvm_cpu_has_pending_timer(vcpu))
1596 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001597 if (signal_pending(current))
1598 break;
1599
Eddie Dongb6958ce2007-07-18 12:15:21 +03001600 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001601 }
1602
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001603 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001604}
1605
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001606#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001607/*
1608 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1609 */
1610void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1611{
1612 int me;
1613 int cpu = vcpu->cpu;
1614 wait_queue_head_t *wqp;
1615
1616 wqp = kvm_arch_vcpu_wq(vcpu);
1617 if (waitqueue_active(wqp)) {
1618 wake_up_interruptible(wqp);
1619 ++vcpu->stat.halt_wakeup;
1620 }
1621
1622 me = get_cpu();
1623 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1624 if (kvm_arch_vcpu_should_kick(vcpu))
1625 smp_send_reschedule(cpu);
1626 put_cpu();
1627}
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001628#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001629
Avi Kivity6aa8b732006-12-10 02:21:36 -08001630void kvm_resched(struct kvm_vcpu *vcpu)
1631{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001632 if (!need_resched())
1633 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001634 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001635}
1636EXPORT_SYMBOL_GPL(kvm_resched);
1637
Konstantin Weitz41628d32012-04-25 15:30:38 +02001638bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1639{
1640 struct pid *pid;
1641 struct task_struct *task = NULL;
1642
1643 rcu_read_lock();
1644 pid = rcu_dereference(target->pid);
1645 if (pid)
1646 task = get_pid_task(target->pid, PIDTYPE_PID);
1647 rcu_read_unlock();
1648 if (!task)
1649 return false;
1650 if (task->flags & PF_VCPU) {
1651 put_task_struct(task);
1652 return false;
1653 }
1654 if (yield_to(task, 1)) {
1655 put_task_struct(task);
1656 return true;
1657 }
1658 put_task_struct(task);
1659 return false;
1660}
1661EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1662
Raghavendra K T06e48c52012-07-19 15:17:52 +05301663#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1664/*
1665 * Helper that checks whether a VCPU is eligible for directed yield.
1666 * Most eligible candidate to yield is decided by following heuristics:
1667 *
1668 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1669 * (preempted lock holder), indicated by @in_spin_loop.
1670 * Set at the beiginning and cleared at the end of interception/PLE handler.
1671 *
1672 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1673 * chance last time (mostly it has become eligible now since we have probably
1674 * yielded to lockholder in last iteration. This is done by toggling
1675 * @dy_eligible each time a VCPU checked for eligibility.)
1676 *
1677 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1678 * to preempted lock-holder could result in wrong VCPU selection and CPU
1679 * burning. Giving priority for a potential lock-holder increases lock
1680 * progress.
1681 *
1682 * Since algorithm is based on heuristics, accessing another VCPU data without
1683 * locking does not harm. It may result in trying to yield to same VCPU, fail
1684 * and continue with next VCPU and so on.
1685 */
1686bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1687{
1688 bool eligible;
1689
1690 eligible = !vcpu->spin_loop.in_spin_loop ||
1691 (vcpu->spin_loop.in_spin_loop &&
1692 vcpu->spin_loop.dy_eligible);
1693
1694 if (vcpu->spin_loop.in_spin_loop)
1695 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1696
1697 return eligible;
1698}
1699#endif
Rik van Riel217ece62011-02-01 09:53:28 -05001700void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001701{
Rik van Riel217ece62011-02-01 09:53:28 -05001702 struct kvm *kvm = me->kvm;
1703 struct kvm_vcpu *vcpu;
1704 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1705 int yielded = 0;
1706 int pass;
1707 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001708
Raghavendra K T4c088492012-07-18 19:07:46 +05301709 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001710 /*
1711 * We boost the priority of a VCPU that is runnable but not
1712 * currently running, because it got preempted by something
1713 * else and called schedule in __vcpu_run. Hopefully that
1714 * VCPU is holding the lock that we need and will release it.
1715 * We approximate round-robin by starting at the last boosted VCPU.
1716 */
1717 for (pass = 0; pass < 2 && !yielded; pass++) {
1718 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001719 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001720 i = last_boosted_vcpu;
1721 continue;
1722 } else if (pass && i > last_boosted_vcpu)
1723 break;
1724 if (vcpu == me)
1725 continue;
1726 if (waitqueue_active(&vcpu->wq))
1727 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301728 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1729 continue;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001730 if (kvm_vcpu_yield_to(vcpu)) {
Rik van Riel217ece62011-02-01 09:53:28 -05001731 kvm->last_boosted_vcpu = i;
1732 yielded = 1;
1733 break;
1734 }
Rik van Riel217ece62011-02-01 09:53:28 -05001735 }
1736 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301737 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301738
1739 /* Ensure vcpu is not eligible during next spinloop */
1740 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001741}
1742EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1743
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001744static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001745{
1746 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001747 struct page *page;
1748
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001749 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001750 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001751#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001752 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001753 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001754#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001755#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1756 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1757 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1758#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001759 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001760 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001761 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001762 vmf->page = page;
1763 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001764}
1765
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001766static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001767 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001768};
1769
1770static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1771{
1772 vma->vm_ops = &kvm_vcpu_vm_ops;
1773 return 0;
1774}
1775
Avi Kivitybccf2152007-02-21 18:04:26 +02001776static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1777{
1778 struct kvm_vcpu *vcpu = filp->private_data;
1779
Al Viro66c0b392008-04-19 20:33:56 +01001780 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001781 return 0;
1782}
1783
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001784static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001785 .release = kvm_vcpu_release,
1786 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001787#ifdef CONFIG_COMPAT
1788 .compat_ioctl = kvm_vcpu_compat_ioctl,
1789#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001790 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001791 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001792};
1793
1794/*
1795 * Allocates an inode for the vcpu.
1796 */
1797static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1798{
Roland Dreier628ff7c2009-12-18 09:41:24 -08001799 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
Avi Kivitybccf2152007-02-21 18:04:26 +02001800}
1801
Avi Kivityc5ea7662007-02-20 18:41:05 +02001802/*
1803 * Creates some virtual cpus. Good luck creating more than one.
1804 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001805static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001806{
1807 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001808 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001809
Gleb Natapov73880c82009-06-09 15:56:28 +03001810 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001811 if (IS_ERR(vcpu))
1812 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001813
Avi Kivity15ad7142007-07-11 18:17:21 +03001814 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1815
Avi Kivity26e52152007-11-20 15:30:24 +02001816 r = kvm_arch_vcpu_setup(vcpu);
1817 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001818 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001819
Shaohua Li11ec2802007-07-23 14:51:37 +08001820 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001821 if (!kvm_vcpu_compatible(vcpu)) {
1822 r = -EINVAL;
1823 goto unlock_vcpu_destroy;
1824 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001825 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1826 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001827 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001828 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001829
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001830 kvm_for_each_vcpu(r, v, kvm)
1831 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001832 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001833 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001834 }
1835
1836 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001837
1838 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001839 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001840 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001841 if (r < 0) {
1842 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001843 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001844 }
1845
1846 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1847 smp_wmb();
1848 atomic_inc(&kvm->online_vcpus);
1849
Gleb Natapov73880c82009-06-09 15:56:28 +03001850 mutex_unlock(&kvm->lock);
Avi Kivitybccf2152007-02-21 18:04:26 +02001851 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001852
Jan Kiszkad7805922011-05-23 10:33:05 +02001853unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001854 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001855vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001856 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001857 return r;
1858}
1859
Avi Kivity1961d272007-03-05 19:46:05 +02001860static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1861{
1862 if (sigset) {
1863 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1864 vcpu->sigset_active = 1;
1865 vcpu->sigset = *sigset;
1866 } else
1867 vcpu->sigset_active = 0;
1868 return 0;
1869}
1870
Avi Kivitybccf2152007-02-21 18:04:26 +02001871static long kvm_vcpu_ioctl(struct file *filp,
1872 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001873{
Avi Kivitybccf2152007-02-21 18:04:26 +02001874 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001875 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001876 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001877 struct kvm_fpu *fpu = NULL;
1878 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001879
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001880 if (vcpu->kvm->mm != current->mm)
1881 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001882
1883#if defined(CONFIG_S390) || defined(CONFIG_PPC)
1884 /*
1885 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1886 * so vcpu_load() would break it.
1887 */
1888 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1889 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1890#endif
1891
1892
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001893 r = vcpu_load(vcpu);
1894 if (r)
1895 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001896 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001897 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001898 r = -EINVAL;
1899 if (arg)
1900 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001901 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001902 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001903 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001904 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001905 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001906
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001907 r = -ENOMEM;
1908 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1909 if (!kvm_regs)
1910 goto out;
1911 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001912 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001913 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001914 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001915 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1916 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001917 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001918out_free1:
1919 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001920 break;
1921 }
1922 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001923 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001924
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001925 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001926 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1927 if (IS_ERR(kvm_regs)) {
1928 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001929 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001930 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001931 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001932 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001933 goto out_free2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001934 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001935out_free2:
1936 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001937 break;
1938 }
1939 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07001940 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1941 r = -ENOMEM;
1942 if (!kvm_sregs)
1943 goto out;
1944 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001945 if (r)
1946 goto out;
1947 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001948 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001949 goto out;
1950 r = 0;
1951 break;
1952 }
1953 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02001954 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1955 if (IS_ERR(kvm_sregs)) {
1956 r = PTR_ERR(kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001957 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001958 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07001959 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001960 if (r)
1961 goto out;
1962 r = 0;
1963 break;
1964 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03001965 case KVM_GET_MP_STATE: {
1966 struct kvm_mp_state mp_state;
1967
1968 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1969 if (r)
1970 goto out;
1971 r = -EFAULT;
1972 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1973 goto out;
1974 r = 0;
1975 break;
1976 }
1977 case KVM_SET_MP_STATE: {
1978 struct kvm_mp_state mp_state;
1979
1980 r = -EFAULT;
1981 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1982 goto out;
1983 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1984 if (r)
1985 goto out;
1986 r = 0;
1987 break;
1988 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001989 case KVM_TRANSLATE: {
1990 struct kvm_translation tr;
1991
1992 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00001993 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08001995 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001996 if (r)
1997 goto out;
1998 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00001999 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002000 goto out;
2001 r = 0;
2002 break;
2003 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002004 case KVM_SET_GUEST_DEBUG: {
2005 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002006
2007 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002008 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002009 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002010 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002011 if (r)
2012 goto out;
2013 r = 0;
2014 break;
2015 }
Avi Kivity1961d272007-03-05 19:46:05 +02002016 case KVM_SET_SIGNAL_MASK: {
2017 struct kvm_signal_mask __user *sigmask_arg = argp;
2018 struct kvm_signal_mask kvm_sigmask;
2019 sigset_t sigset, *p;
2020
2021 p = NULL;
2022 if (argp) {
2023 r = -EFAULT;
2024 if (copy_from_user(&kvm_sigmask, argp,
2025 sizeof kvm_sigmask))
2026 goto out;
2027 r = -EINVAL;
2028 if (kvm_sigmask.len != sizeof sigset)
2029 goto out;
2030 r = -EFAULT;
2031 if (copy_from_user(&sigset, sigmask_arg->sigset,
2032 sizeof sigset))
2033 goto out;
2034 p = &sigset;
2035 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002036 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002037 break;
2038 }
Avi Kivityb8836732007-04-01 16:34:31 +03002039 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002040 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2041 r = -ENOMEM;
2042 if (!fpu)
2043 goto out;
2044 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002045 if (r)
2046 goto out;
2047 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002048 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002049 goto out;
2050 r = 0;
2051 break;
2052 }
2053 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002054 fpu = memdup_user(argp, sizeof(*fpu));
2055 if (IS_ERR(fpu)) {
2056 r = PTR_ERR(fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002057 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002058 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002059 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002060 if (r)
2061 goto out;
2062 r = 0;
2063 break;
2064 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002065 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002066 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002067 }
2068out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002069 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002070 kfree(fpu);
2071 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002072 return r;
2073}
2074
Alexander Graf1dda6062011-06-08 02:45:37 +02002075#ifdef CONFIG_COMPAT
2076static long kvm_vcpu_compat_ioctl(struct file *filp,
2077 unsigned int ioctl, unsigned long arg)
2078{
2079 struct kvm_vcpu *vcpu = filp->private_data;
2080 void __user *argp = compat_ptr(arg);
2081 int r;
2082
2083 if (vcpu->kvm->mm != current->mm)
2084 return -EIO;
2085
2086 switch (ioctl) {
2087 case KVM_SET_SIGNAL_MASK: {
2088 struct kvm_signal_mask __user *sigmask_arg = argp;
2089 struct kvm_signal_mask kvm_sigmask;
2090 compat_sigset_t csigset;
2091 sigset_t sigset;
2092
2093 if (argp) {
2094 r = -EFAULT;
2095 if (copy_from_user(&kvm_sigmask, argp,
2096 sizeof kvm_sigmask))
2097 goto out;
2098 r = -EINVAL;
2099 if (kvm_sigmask.len != sizeof csigset)
2100 goto out;
2101 r = -EFAULT;
2102 if (copy_from_user(&csigset, sigmask_arg->sigset,
2103 sizeof csigset))
2104 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002105 sigset_from_compat(&sigset, &csigset);
2106 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2107 } else
2108 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002109 break;
2110 }
2111 default:
2112 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2113 }
2114
2115out:
2116 return r;
2117}
2118#endif
2119
Avi Kivitybccf2152007-02-21 18:04:26 +02002120static long kvm_vm_ioctl(struct file *filp,
2121 unsigned int ioctl, unsigned long arg)
2122{
2123 struct kvm *kvm = filp->private_data;
2124 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002125 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002126
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002127 if (kvm->mm != current->mm)
2128 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002129 switch (ioctl) {
2130 case KVM_CREATE_VCPU:
2131 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2132 if (r < 0)
2133 goto out;
2134 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002135 case KVM_SET_USER_MEMORY_REGION: {
2136 struct kvm_userspace_memory_region kvm_userspace_mem;
2137
2138 r = -EFAULT;
2139 if (copy_from_user(&kvm_userspace_mem, argp,
2140 sizeof kvm_userspace_mem))
2141 goto out;
2142
2143 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002144 if (r)
2145 goto out;
2146 break;
2147 }
2148 case KVM_GET_DIRTY_LOG: {
2149 struct kvm_dirty_log log;
2150
2151 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002152 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002153 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002154 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002155 if (r)
2156 goto out;
2157 break;
2158 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002159#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2160 case KVM_REGISTER_COALESCED_MMIO: {
2161 struct kvm_coalesced_mmio_zone zone;
2162 r = -EFAULT;
2163 if (copy_from_user(&zone, argp, sizeof zone))
2164 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002165 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2166 if (r)
2167 goto out;
2168 r = 0;
2169 break;
2170 }
2171 case KVM_UNREGISTER_COALESCED_MMIO: {
2172 struct kvm_coalesced_mmio_zone zone;
2173 r = -EFAULT;
2174 if (copy_from_user(&zone, argp, sizeof zone))
2175 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002176 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2177 if (r)
2178 goto out;
2179 r = 0;
2180 break;
2181 }
2182#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002183 case KVM_IRQFD: {
2184 struct kvm_irqfd data;
2185
2186 r = -EFAULT;
2187 if (copy_from_user(&data, argp, sizeof data))
2188 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002189 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002190 break;
2191 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002192 case KVM_IOEVENTFD: {
2193 struct kvm_ioeventfd data;
2194
2195 r = -EFAULT;
2196 if (copy_from_user(&data, argp, sizeof data))
2197 goto out;
2198 r = kvm_ioeventfd(kvm, &data);
2199 break;
2200 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002201#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2202 case KVM_SET_BOOT_CPU_ID:
2203 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002204 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002205 if (atomic_read(&kvm->online_vcpus) != 0)
2206 r = -EBUSY;
2207 else
2208 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002209 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002210 break;
2211#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002212#ifdef CONFIG_HAVE_KVM_MSI
2213 case KVM_SIGNAL_MSI: {
2214 struct kvm_msi msi;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&msi, argp, sizeof msi))
2218 goto out;
2219 r = kvm_send_userspace_msi(kvm, &msi);
2220 break;
2221 }
2222#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002223#ifdef __KVM_HAVE_IRQ_LINE
2224 case KVM_IRQ_LINE_STATUS:
2225 case KVM_IRQ_LINE: {
2226 struct kvm_irq_level irq_event;
2227
2228 r = -EFAULT;
2229 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2230 goto out;
2231
2232 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2233 if (r)
2234 goto out;
2235
2236 r = -EFAULT;
2237 if (ioctl == KVM_IRQ_LINE_STATUS) {
2238 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2239 goto out;
2240 }
2241
2242 r = 0;
2243 break;
2244 }
2245#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002246 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002247 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002248 if (r == -ENOTTY)
2249 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002250 }
2251out:
2252 return r;
2253}
2254
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002255#ifdef CONFIG_COMPAT
2256struct compat_kvm_dirty_log {
2257 __u32 slot;
2258 __u32 padding1;
2259 union {
2260 compat_uptr_t dirty_bitmap; /* one bit per page */
2261 __u64 padding2;
2262 };
2263};
2264
2265static long kvm_vm_compat_ioctl(struct file *filp,
2266 unsigned int ioctl, unsigned long arg)
2267{
2268 struct kvm *kvm = filp->private_data;
2269 int r;
2270
2271 if (kvm->mm != current->mm)
2272 return -EIO;
2273 switch (ioctl) {
2274 case KVM_GET_DIRTY_LOG: {
2275 struct compat_kvm_dirty_log compat_log;
2276 struct kvm_dirty_log log;
2277
2278 r = -EFAULT;
2279 if (copy_from_user(&compat_log, (void __user *)arg,
2280 sizeof(compat_log)))
2281 goto out;
2282 log.slot = compat_log.slot;
2283 log.padding1 = compat_log.padding1;
2284 log.padding2 = compat_log.padding2;
2285 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2286
2287 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2288 if (r)
2289 goto out;
2290 break;
2291 }
2292 default:
2293 r = kvm_vm_ioctl(filp, ioctl, arg);
2294 }
2295
2296out:
2297 return r;
2298}
2299#endif
2300
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002301static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002302{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002303 struct page *page[1];
2304 unsigned long addr;
2305 int npages;
2306 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002307 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002308
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002309 addr = gfn_to_hva(kvm, gfn);
2310 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002311 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002312
2313 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2314 NULL);
2315 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002316 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002317
2318 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002319 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002320}
2321
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002322static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002323 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002324};
2325
2326static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2327{
2328 vma->vm_ops = &kvm_vm_vm_ops;
2329 return 0;
2330}
2331
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002332static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002333 .release = kvm_vm_release,
2334 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002335#ifdef CONFIG_COMPAT
2336 .compat_ioctl = kvm_vm_compat_ioctl,
2337#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002338 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002339 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002340};
2341
Carsten Ottee08b9632012-01-04 10:25:20 +01002342static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002343{
Heiko Carstensaac87632010-10-27 17:22:10 +02002344 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002345 struct kvm *kvm;
2346
Carsten Ottee08b9632012-01-04 10:25:20 +01002347 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002348 if (IS_ERR(kvm))
2349 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002350#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2351 r = kvm_coalesced_mmio_init(kvm);
2352 if (r < 0) {
2353 kvm_put_kvm(kvm);
2354 return r;
2355 }
2356#endif
Heiko Carstensaac87632010-10-27 17:22:10 +02002357 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2358 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002359 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002360
Heiko Carstensaac87632010-10-27 17:22:10 +02002361 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002362}
2363
Avi Kivity1a811b62008-12-08 18:25:27 +02002364static long kvm_dev_ioctl_check_extension_generic(long arg)
2365{
2366 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002367 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002368 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002369 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002370#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2371 case KVM_CAP_SET_BOOT_CPU_ID:
2372#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002373 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002374#ifdef CONFIG_HAVE_KVM_MSI
2375 case KVM_CAP_SIGNAL_MSI:
2376#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002377 return 1;
Marc Zyngier9900b4b2012-06-15 15:07:02 -04002378#ifdef KVM_CAP_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002379 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002380 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002381#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002382 default:
2383 break;
2384 }
2385 return kvm_dev_ioctl_check_extension(arg);
2386}
2387
Avi Kivityf17abe92007-02-21 19:28:04 +02002388static long kvm_dev_ioctl(struct file *filp,
2389 unsigned int ioctl, unsigned long arg)
2390{
Avi Kivity07c45a32007-03-07 13:05:38 +02002391 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002392
2393 switch (ioctl) {
2394 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002395 r = -EINVAL;
2396 if (arg)
2397 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002398 r = KVM_API_VERSION;
2399 break;
2400 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002401 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002402 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002403 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002404 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002405 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002406 case KVM_GET_VCPU_MMAP_SIZE:
2407 r = -EINVAL;
2408 if (arg)
2409 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002410 r = PAGE_SIZE; /* struct kvm_run */
2411#ifdef CONFIG_X86
2412 r += PAGE_SIZE; /* pio data page */
2413#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002414#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2415 r += PAGE_SIZE; /* coalesced mmio ring page */
2416#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002417 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002418 case KVM_TRACE_ENABLE:
2419 case KVM_TRACE_PAUSE:
2420 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002421 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002422 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002423 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002424 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002425 }
2426out:
2427 return r;
2428}
2429
Avi Kivity6aa8b732006-12-10 02:21:36 -08002430static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002431 .unlocked_ioctl = kvm_dev_ioctl,
2432 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002433 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002434};
2435
2436static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002437 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002438 "kvm",
2439 &kvm_chardev_ops,
2440};
2441
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002442static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002443{
2444 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002445 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002446
Rusty Russell7f59f492008-12-07 21:25:45 +10302447 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002448 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002449
Rusty Russell7f59f492008-12-07 21:25:45 +10302450 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002451
2452 r = kvm_arch_hardware_enable(NULL);
2453
2454 if (r) {
2455 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2456 atomic_inc(&hardware_enable_failed);
2457 printk(KERN_INFO "kvm: enabling virtualization on "
2458 "CPU%d failed\n", cpu);
2459 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002460}
2461
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002462static void hardware_enable(void *junk)
2463{
Jan Kiszkae935b832011-02-08 12:55:33 +01002464 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002465 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002466 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002467}
2468
2469static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002470{
2471 int cpu = raw_smp_processor_id();
2472
Rusty Russell7f59f492008-12-07 21:25:45 +10302473 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002474 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302475 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002476 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002477}
2478
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002479static void hardware_disable(void *junk)
2480{
Jan Kiszkae935b832011-02-08 12:55:33 +01002481 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002482 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002483 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002484}
2485
Alexander Graf10474ae2009-09-15 11:37:46 +02002486static void hardware_disable_all_nolock(void)
2487{
2488 BUG_ON(!kvm_usage_count);
2489
2490 kvm_usage_count--;
2491 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002492 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002493}
2494
2495static void hardware_disable_all(void)
2496{
Jan Kiszkae935b832011-02-08 12:55:33 +01002497 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002498 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002499 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002500}
2501
2502static int hardware_enable_all(void)
2503{
2504 int r = 0;
2505
Jan Kiszkae935b832011-02-08 12:55:33 +01002506 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002507
2508 kvm_usage_count++;
2509 if (kvm_usage_count == 1) {
2510 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002511 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002512
2513 if (atomic_read(&hardware_enable_failed)) {
2514 hardware_disable_all_nolock();
2515 r = -EBUSY;
2516 }
2517 }
2518
Jan Kiszkae935b832011-02-08 12:55:33 +01002519 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002520
2521 return r;
2522}
2523
Avi Kivity774c47f2007-02-12 00:54:47 -08002524static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2525 void *v)
2526{
2527 int cpu = (long)v;
2528
Alexander Graf10474ae2009-09-15 11:37:46 +02002529 if (!kvm_usage_count)
2530 return NOTIFY_OK;
2531
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002532 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002533 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002534 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002535 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2536 cpu);
2537 hardware_disable(NULL);
2538 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002539 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002540 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2541 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002542 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002543 break;
2544 }
2545 return NOTIFY_OK;
2546}
2547
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002548
Avi Kivityb7c41452010-12-02 17:52:50 +02002549asmlinkage void kvm_spurious_fault(void)
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002550{
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002551 /* Fault while not rebooting. We want the trace. */
2552 BUG();
2553}
Avi Kivityb7c41452010-12-02 17:52:50 +02002554EXPORT_SYMBOL_GPL(kvm_spurious_fault);
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002555
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002556static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002557 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002558{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002559 /*
2560 * Some (well, at least mine) BIOSes hang on reboot if
2561 * in vmx root mode.
2562 *
2563 * And Intel TXT required VMX off for all cpu when system shutdown.
2564 */
2565 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2566 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002567 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002568 return NOTIFY_OK;
2569}
2570
2571static struct notifier_block kvm_reboot_notifier = {
2572 .notifier_call = kvm_reboot,
2573 .priority = 0,
2574};
2575
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002576static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002577{
2578 int i;
2579
2580 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002581 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002582
2583 kvm_iodevice_destructor(pos);
2584 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002585 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002586}
2587
Sasha Levin743eeb02011-07-27 16:00:48 +03002588int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2589{
2590 const struct kvm_io_range *r1 = p1;
2591 const struct kvm_io_range *r2 = p2;
2592
2593 if (r1->addr < r2->addr)
2594 return -1;
2595 if (r1->addr + r1->len > r2->addr + r2->len)
2596 return 1;
2597 return 0;
2598}
2599
2600int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2601 gpa_t addr, int len)
2602{
Sasha Levin743eeb02011-07-27 16:00:48 +03002603 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2604 .addr = addr,
2605 .len = len,
2606 .dev = dev,
2607 };
2608
2609 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2610 kvm_io_bus_sort_cmp, NULL);
2611
2612 return 0;
2613}
2614
2615int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2616 gpa_t addr, int len)
2617{
2618 struct kvm_io_range *range, key;
2619 int off;
2620
2621 key = (struct kvm_io_range) {
2622 .addr = addr,
2623 .len = len,
2624 };
2625
2626 range = bsearch(&key, bus->range, bus->dev_count,
2627 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2628 if (range == NULL)
2629 return -ENOENT;
2630
2631 off = range - bus->range;
2632
2633 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2634 off--;
2635
2636 return off;
2637}
2638
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002639/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002640int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002641 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002642{
Sasha Levin743eeb02011-07-27 16:00:48 +03002643 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002644 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002645 struct kvm_io_range range;
2646
2647 range = (struct kvm_io_range) {
2648 .addr = addr,
2649 .len = len,
2650 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002651
2652 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002653 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2654 if (idx < 0)
2655 return -EOPNOTSUPP;
2656
2657 while (idx < bus->dev_count &&
2658 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2659 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002660 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002661 idx++;
2662 }
2663
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002664 return -EOPNOTSUPP;
2665}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002666
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002667/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002668int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2669 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002670{
Sasha Levin743eeb02011-07-27 16:00:48 +03002671 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002672 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002673 struct kvm_io_range range;
2674
2675 range = (struct kvm_io_range) {
2676 .addr = addr,
2677 .len = len,
2678 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002679
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002680 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002681 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2682 if (idx < 0)
2683 return -EOPNOTSUPP;
2684
2685 while (idx < bus->dev_count &&
2686 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2687 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002688 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002689 idx++;
2690 }
2691
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002692 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002693}
2694
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002695/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002696int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2697 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002698{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002699 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002700
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002701 bus = kvm->buses[bus_idx];
Amos Konga13007162012-03-09 12:17:32 +08002702 if (bus->dev_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002703 return -ENOSPC;
2704
Amos Konga13007162012-03-09 12:17:32 +08002705 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2706 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002707 if (!new_bus)
2708 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002709 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2710 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002711 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002712 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2713 synchronize_srcu_expedited(&kvm->srcu);
2714 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002715
2716 return 0;
2717}
2718
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002719/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002720int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2721 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002722{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002723 int i, r;
2724 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002725
Sasha Levincdfca7b2011-12-04 19:36:28 +02002726 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002727 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002728 for (i = 0; i < bus->dev_count; i++)
2729 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002730 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002731 break;
2732 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002733
Amos Konga13007162012-03-09 12:17:32 +08002734 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002735 return r;
Amos Konga13007162012-03-09 12:17:32 +08002736
2737 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2738 sizeof(struct kvm_io_range)), GFP_KERNEL);
2739 if (!new_bus)
2740 return -ENOMEM;
2741
2742 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2743 new_bus->dev_count--;
2744 memcpy(new_bus->range + i, bus->range + i + 1,
2745 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002746
2747 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2748 synchronize_srcu_expedited(&kvm->srcu);
2749 kfree(bus);
2750 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002751}
2752
Avi Kivity774c47f2007-02-12 00:54:47 -08002753static struct notifier_block kvm_cpu_notifier = {
2754 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08002755};
2756
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002757static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002758{
2759 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002760 struct kvm *kvm;
2761
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002762 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002763 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02002764 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002765 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01002766 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002767 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002768}
2769
2770DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2771
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002772static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03002773{
2774 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002775 struct kvm *kvm;
2776 struct kvm_vcpu *vcpu;
2777 int i;
2778
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002779 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002780 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03002781 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002782 kvm_for_each_vcpu(i, vcpu, kvm)
2783 *val += *(u32 *)((void *)vcpu + offset);
2784
Jan Kiszkae935b832011-02-08 12:55:33 +01002785 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002786 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002787}
2788
Avi Kivityba1389b2007-11-18 16:24:12 +02002789DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2790
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002791static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02002792 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2793 [KVM_STAT_VM] = &vm_stat_fops,
2794};
Avi Kivity1165f5f2007-04-19 17:27:43 +03002795
Hamo4f69b682011-12-15 14:23:16 +08002796static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002797{
Hamo4f69b682011-12-15 14:23:16 +08002798 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002799 struct kvm_stats_debugfs_item *p;
2800
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002801 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08002802 if (kvm_debugfs_dir == NULL)
2803 goto out;
2804
2805 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002806 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03002807 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02002808 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08002809 if (p->dentry == NULL)
2810 goto out_dir;
2811 }
2812
2813 return 0;
2814
2815out_dir:
2816 debugfs_remove_recursive(kvm_debugfs_dir);
2817out:
2818 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002819}
2820
2821static void kvm_exit_debug(void)
2822{
2823 struct kvm_stats_debugfs_item *p;
2824
2825 for (p = debugfs_entries; p->name; ++p)
2826 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002827 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002828}
2829
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002830static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002831{
Alexander Graf10474ae2009-09-15 11:37:46 +02002832 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002833 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002834 return 0;
2835}
2836
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002837static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002838{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002839 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01002840 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002841 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002842 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08002843}
2844
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002845static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08002846 .suspend = kvm_suspend,
2847 .resume = kvm_resume,
2848};
2849
Avi Kivity15ad7142007-07-11 18:17:21 +03002850static inline
2851struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2852{
2853 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2854}
2855
2856static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2857{
2858 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2859
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002860 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002861}
2862
2863static void kvm_sched_out(struct preempt_notifier *pn,
2864 struct task_struct *next)
2865{
2866 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2867
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002868 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002869}
2870
Avi Kivity0ee75be2010-04-28 15:39:01 +03002871int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10002872 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002873{
2874 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002875 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002876
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002877 r = kvm_arch_init(opaque);
2878 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002879 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002880
Avi Kivity8437a612009-06-06 14:52:35 -07002881 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10302882 r = -ENOMEM;
2883 goto out_free_0;
2884 }
2885
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002886 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002887 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10302888 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002889
Yang, Sheng002c7f72007-07-31 14:23:01 +03002890 for_each_online_cpu(cpu) {
2891 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002892 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02002893 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03002894 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002895 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002896 }
2897
Avi Kivity774c47f2007-02-12 00:54:47 -08002898 r = register_cpu_notifier(&kvm_cpu_notifier);
2899 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002900 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002901 register_reboot_notifier(&kvm_reboot_notifier);
2902
Rusty Russellc16f8622007-07-30 21:12:19 +10002903 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03002904 if (!vcpu_align)
2905 vcpu_align = __alignof__(struct kvm_vcpu);
2906 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08002907 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10002908 if (!kvm_vcpu_cache) {
2909 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002910 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10002911 }
2912
Gleb Natapovaf585b92010-10-14 11:22:46 +02002913 r = kvm_async_pf_init();
2914 if (r)
2915 goto out_free;
2916
Avi Kivity6aa8b732006-12-10 02:21:36 -08002917 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002918 kvm_vm_fops.owner = module;
2919 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002920
2921 r = misc_register(&kvm_dev);
2922 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04002923 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02002924 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002925 }
2926
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002927 register_syscore_ops(&kvm_syscore_ops);
2928
Avi Kivity15ad7142007-07-11 18:17:21 +03002929 kvm_preempt_ops.sched_in = kvm_sched_in;
2930 kvm_preempt_ops.sched_out = kvm_sched_out;
2931
Hamo4f69b682011-12-15 14:23:16 +08002932 r = kvm_init_debug();
2933 if (r) {
2934 printk(KERN_ERR "kvm: create debugfs files failed\n");
2935 goto out_undebugfs;
2936 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07002937
Avi Kivityc7addb92007-09-16 18:58:32 +02002938 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002939
Hamo4f69b682011-12-15 14:23:16 +08002940out_undebugfs:
2941 unregister_syscore_ops(&kvm_syscore_ops);
Gleb Natapovaf585b92010-10-14 11:22:46 +02002942out_unreg:
2943 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002944out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10002945 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002946out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002947 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08002948 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002949out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08002950out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002951 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10302952out_free_0a:
2953 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002954out_free_0:
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002955 kvm_arch_exit();
Zhang Xiantaod23087842007-11-29 15:35:39 +08002956out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002957 return r;
2958}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002959EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002960
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002961void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002962{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07002963 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002964 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10002965 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02002966 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002967 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002968 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002969 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002970 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002971 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002972 kvm_arch_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10302973 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002974}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002975EXPORT_SYMBOL_GPL(kvm_exit);