blob: 1cd693a76a510c34a2d563d9fe68072802d28620 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Jan Kiszkae935b832011-02-08 12:55:33 +010073DEFINE_RAW_SPINLOCK(kvm_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080075
Rusty Russell7f59f492008-12-07 21:25:45 +103076static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020077static int kvm_usage_count = 0;
78static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030079
Rusty Russellc16f8622007-07-30 21:12:19 +100080struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030082
Avi Kivity15ad7142007-07-11 18:17:21 +030083static __read_mostly struct preempt_ops kvm_preempt_ops;
84
Hollis Blanchard76f7c872008-04-15 16:05:42 -050085struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080086
Avi Kivitybccf2152007-02-21 18:04:26 +020087static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020089#ifdef CONFIG_COMPAT
90static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020093static int hardware_enable_all(void);
94static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020095
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020096static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
Avi Kivityb7c41452010-12-02 17:52:50 +020098bool kvm_rebooting;
99EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300100
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300101static bool largepages_enabled = true;
102
Xiao Guangronga2766322012-07-26 11:58:59 +0800103bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300104{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100105 if (pfn_valid(pfn)) {
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800106 int reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800107 struct page *tail = pfn_to_page(pfn);
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800110 if (head != tail) {
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800111 /*
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
119 * what we just read.
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800120 */
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800121 smp_rmb();
122 if (PageTail(tail))
123 return reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800124 }
125 return PageReserved(tail);
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100126 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300127
128 return true;
129}
130
Avi Kivity6aa8b732006-12-10 02:21:36 -0800131/*
132 * Switches to specified vcpu, until a matching vcpu_put()
133 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300134int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135{
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 int cpu;
137
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300138 if (mutex_lock_killable(&vcpu->mutex))
139 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
145 synchronize_rcu();
146 put_pid(oldpid);
147 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300148 cpu = get_cpu();
149 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200150 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300151 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300152 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200153}
154
Carsten Otte313a3dc2007-10-11 19:16:52 +0200155void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800156{
Avi Kivity15ad7142007-07-11 18:17:21 +0300157 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200158 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300159 preempt_notifier_unregister(&vcpu->preempt_notifier);
160 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800161 mutex_unlock(&vcpu->mutex);
162}
163
Avi Kivityd9e368d2007-06-07 19:18:30 +0300164static void ack_flush(void *_completed)
165{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300166}
167
Rusty Russell49846892008-12-08 20:26:24 +1030168static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300169{
Avi Kivity597a5f52008-07-20 14:24:22 +0300170 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030171 cpumask_var_t cpus;
172 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300173 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300174
Li Zefan79f55992009-06-15 14:58:26 +0800175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800177 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300178 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800179 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300180 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800181
182 /* Set ->requests bit before we read ->mode */
183 smp_mb();
184
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030187 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300188 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
193 else
194 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800195 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030196 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030197 return called;
198}
199
200void kvm_flush_remote_tlbs(struct kvm *kvm)
201{
Alex Shibec87d62012-03-04 13:13:38 +0800202 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800203
204 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300208}
209
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500210void kvm_reload_remote_mmus(struct kvm *kvm)
211{
Rusty Russell49846892008-12-08 20:26:24 +1030212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500213}
214
Marcelo Tosattid8281992012-11-27 23:29:01 -0200215void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216{
217 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218}
219
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000220int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
221{
222 struct page *page;
223 int r;
224
225 mutex_init(&vcpu->mutex);
226 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000227 vcpu->kvm = kvm;
228 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500229 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300230 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200231 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232
233 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
234 if (!page) {
235 r = -ENOMEM;
236 goto fail;
237 }
238 vcpu->run = page_address(page);
239
Raghavendra K T4c088492012-07-18 19:07:46 +0530240 kvm_vcpu_set_in_spin_loop(vcpu, false);
241 kvm_vcpu_set_dy_eligible(vcpu, false);
242
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800243 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000244 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800245 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000246 return 0;
247
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000248fail_free_run:
249 free_page((unsigned long)vcpu->run);
250fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000251 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000252}
253EXPORT_SYMBOL_GPL(kvm_vcpu_init);
254
255void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
256{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500257 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800258 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000259 free_page((unsigned long)vcpu->run);
260}
261EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
262
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200263#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
265{
266 return container_of(mn, struct kvm, mmu_notifier);
267}
268
269static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
270 struct mm_struct *mm,
271 unsigned long address)
272{
273 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200274 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200275
276 /*
277 * When ->invalidate_page runs, the linux pte has been zapped
278 * already but the page is still allocated until
279 * ->invalidate_page returns. So if we increase the sequence
280 * here the kvm page fault will notice if the spte can't be
281 * established because the page is going to be freed. If
282 * instead the kvm page fault establishes the spte before
283 * ->invalidate_page runs, kvm_unmap_hva will release it
284 * before returning.
285 *
286 * The sequence increase only need to be seen at spin_unlock
287 * time, and not at spin_lock time.
288 *
289 * Increasing the sequence after the spin_unlock would be
290 * unsafe because the kvm page fault could then establish the
291 * pte after kvm_unmap_hva returned, without noticing the page
292 * is going to be freed.
293 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200294 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200295 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900296
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200297 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800298 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200299 /* we've to flush the tlb before the pages can be freed */
300 if (need_tlb_flush)
301 kvm_flush_remote_tlbs(kvm);
302
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900303 spin_unlock(&kvm->mmu_lock);
304 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200305}
306
Izik Eidus3da0dd42009-09-23 21:47:18 +0300307static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
308 struct mm_struct *mm,
309 unsigned long address,
310 pte_t pte)
311{
312 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200313 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300314
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200315 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300316 spin_lock(&kvm->mmu_lock);
317 kvm->mmu_notifier_seq++;
318 kvm_set_spte_hva(kvm, address, pte);
319 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200320 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300321}
322
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200323static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
324 struct mm_struct *mm,
325 unsigned long start,
326 unsigned long end)
327{
328 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200329 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200330
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200331 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200332 spin_lock(&kvm->mmu_lock);
333 /*
334 * The count increase must become visible at unlock time as no
335 * spte can be established without taking the mmu_lock and
336 * count is also read inside the mmu_lock critical section.
337 */
338 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900339 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800340 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200341 /* we've to flush the tlb before the pages can be freed */
342 if (need_tlb_flush)
343 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900344
345 spin_unlock(&kvm->mmu_lock);
346 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200347}
348
349static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
350 struct mm_struct *mm,
351 unsigned long start,
352 unsigned long end)
353{
354 struct kvm *kvm = mmu_notifier_to_kvm(mn);
355
356 spin_lock(&kvm->mmu_lock);
357 /*
358 * This sequence increase will notify the kvm page fault that
359 * the page that is going to be mapped in the spte could have
360 * been freed.
361 */
362 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000363 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200364 /*
365 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000366 * below count decrease, which is ensured by the smp_wmb above
367 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200368 */
369 kvm->mmu_notifier_count--;
370 spin_unlock(&kvm->mmu_lock);
371
372 BUG_ON(kvm->mmu_notifier_count < 0);
373}
374
375static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
376 struct mm_struct *mm,
377 unsigned long address)
378{
379 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200380 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200381
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200382 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200383 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200384
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900385 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200386 if (young)
387 kvm_flush_remote_tlbs(kvm);
388
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900389 spin_unlock(&kvm->mmu_lock);
390 srcu_read_unlock(&kvm->srcu, idx);
391
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200392 return young;
393}
394
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800395static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
396 struct mm_struct *mm,
397 unsigned long address)
398{
399 struct kvm *kvm = mmu_notifier_to_kvm(mn);
400 int young, idx;
401
402 idx = srcu_read_lock(&kvm->srcu);
403 spin_lock(&kvm->mmu_lock);
404 young = kvm_test_age_hva(kvm, address);
405 spin_unlock(&kvm->mmu_lock);
406 srcu_read_unlock(&kvm->srcu, idx);
407
408 return young;
409}
410
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100411static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
412 struct mm_struct *mm)
413{
414 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800415 int idx;
416
417 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300418 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800419 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100420}
421
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200422static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
423 .invalidate_page = kvm_mmu_notifier_invalidate_page,
424 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
425 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
426 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800427 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300428 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100429 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200430};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200431
432static int kvm_init_mmu_notifier(struct kvm *kvm)
433{
434 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
435 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
436}
437
438#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439
440static int kvm_init_mmu_notifier(struct kvm *kvm)
441{
442 return 0;
443}
444
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200445#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800447static void kvm_init_memslots_id(struct kvm *kvm)
448{
449 int i;
450 struct kvm_memslots *slots = kvm->memslots;
451
452 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800453 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800454}
455
Carsten Ottee08b9632012-01-04 10:25:20 +0100456static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800457{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100458 int r, i;
459 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800460
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100461 if (!kvm)
462 return ERR_PTR(-ENOMEM);
463
Carsten Ottee08b9632012-01-04 10:25:20 +0100464 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100465 if (r)
466 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200467
468 r = hardware_enable_all();
469 if (r)
470 goto out_err_nodisable;
471
Avi Kivity75858a82009-01-04 17:10:50 +0200472#ifdef CONFIG_HAVE_KVM_IRQCHIP
473 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300474 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200475#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800476
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200477 r = -ENOMEM;
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
479 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100480 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800481 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200482 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100483 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
486 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100487 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200488 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200489 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700490
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900491 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400494 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800495 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300496 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200497 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300498 atomic_set(&kvm->users_count, 1);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700499
500 r = kvm_init_mmu_notifier(kvm);
501 if (r)
502 goto out_err;
503
Jan Kiszkae935b832011-02-08 12:55:33 +0100504 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000505 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100506 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100507
Avi Kivityf17abe92007-02-21 19:28:04 +0200508 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200509
510out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100511 cleanup_srcu_struct(&kvm->srcu);
512out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200513 hardware_disable_all();
514out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200515 for (i = 0; i < KVM_NR_BUSES; i++)
516 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200517 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100518 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200519 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200520}
521
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900522/*
523 * Avoid using vmalloc for a small buffer.
524 * Should not be used when the size is statically known.
525 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900526void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900527{
528 if (size > PAGE_SIZE)
529 return vzalloc(size);
530 else
531 return kzalloc(size, GFP_KERNEL);
532}
533
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900534void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900535{
536 if (is_vmalloc_addr(addr))
537 vfree(addr);
538 else
539 kfree(addr);
540}
541
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900542static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
543{
544 if (!memslot->dirty_bitmap)
545 return;
546
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900547 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900548 memslot->dirty_bitmap = NULL;
549}
550
Avi Kivity6aa8b732006-12-10 02:21:36 -0800551/*
552 * Free any memory in @free but not in @dont.
553 */
554static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
555 struct kvm_memory_slot *dont)
556{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900558 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800559
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900560 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300561
Avi Kivity6aa8b732006-12-10 02:21:36 -0800562 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563}
564
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800565void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200567 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800568 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200572
573 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800574}
575
Avi Kivityf17abe92007-02-21 19:28:04 +0200576static void kvm_destroy_vm(struct kvm *kvm)
577{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200578 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200579 struct mm_struct *mm = kvm->mm;
580
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800581 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100582 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800583 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100584 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200585 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200586 for (i = 0; i < KVM_NR_BUSES; i++)
587 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200588 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200589#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200591#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300592 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200593#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800594 kvm_arch_destroy_vm(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100595 kvm_free_physmem(kvm);
596 cleanup_srcu_struct(&kvm->srcu);
597 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200598 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200599 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200600}
601
Izik Eidusd39f13b2008-03-30 16:01:25 +0300602void kvm_get_kvm(struct kvm *kvm)
603{
604 atomic_inc(&kvm->users_count);
605}
606EXPORT_SYMBOL_GPL(kvm_get_kvm);
607
608void kvm_put_kvm(struct kvm *kvm)
609{
610 if (atomic_dec_and_test(&kvm->users_count))
611 kvm_destroy_vm(kvm);
612}
613EXPORT_SYMBOL_GPL(kvm_put_kvm);
614
615
Avi Kivityf17abe92007-02-21 19:28:04 +0200616static int kvm_vm_release(struct inode *inode, struct file *filp)
617{
618 struct kvm *kvm = filp->private_data;
619
Gregory Haskins721eecbf2009-05-20 10:30:49 -0400620 kvm_irqfd_release(kvm);
621
Izik Eidusd39f13b2008-03-30 16:01:25 +0300622 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800623 return 0;
624}
625
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900626/*
627 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900628 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900629 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900630static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
631{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900632#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900633 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900634
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900635 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900636 if (!memslot->dirty_bitmap)
637 return -ENOMEM;
638
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900639#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900640 return 0;
641}
642
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800643static int cmp_memslot(const void *slot1, const void *slot2)
644{
645 struct kvm_memory_slot *s1, *s2;
646
647 s1 = (struct kvm_memory_slot *)slot1;
648 s2 = (struct kvm_memory_slot *)slot2;
649
650 if (s1->npages < s2->npages)
651 return 1;
652 if (s1->npages > s2->npages)
653 return -1;
654
655 return 0;
656}
657
658/*
659 * Sort the memslots base on its size, so the larger slots
660 * will get better fit.
661 */
662static void sort_memslots(struct kvm_memslots *slots)
663{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800664 int i;
665
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800666 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
667 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800668
669 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
670 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800671}
672
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800673void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
674{
675 if (new) {
676 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800677 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800678 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800679
Xiao Guangrong28a37542011-11-24 19:04:35 +0800680 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800681 if (new->npages != npages)
682 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800683 }
684
685 slots->generation++;
686}
687
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800688static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
689{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800690 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
691
692#ifdef KVM_CAP_READONLY_MEM
693 valid_flags |= KVM_MEM_READONLY;
694#endif
695
696 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800697 return -EINVAL;
698
699 return 0;
700}
701
Avi Kivity6aa8b732006-12-10 02:21:36 -0800702/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800703 * Allocate some memory and give it an address in the guest physical address
704 * space.
705 *
706 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800707 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500708 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800709 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800710int __kvm_set_memory_region(struct kvm *kvm,
711 struct kvm_userspace_memory_region *mem,
712 int user_alloc)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800713{
Avi Kivity8234b222010-12-27 12:08:45 +0200714 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800715 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200716 unsigned long npages;
Alex Williamson54193692012-11-29 14:07:59 -0700717 struct kvm_memory_slot *memslot, *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800718 struct kvm_memory_slot old, new;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200719 struct kvm_memslots *slots, *old_memslots;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800720
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800721 r = check_memory_region_flags(mem);
722 if (r)
723 goto out;
724
Avi Kivity6aa8b732006-12-10 02:21:36 -0800725 r = -EINVAL;
726 /* General sanity checks */
727 if (mem->memory_size & (PAGE_SIZE - 1))
728 goto out;
729 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
730 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900731 /* We can read the guest memory with __xxx_user() later on. */
732 if (user_alloc &&
733 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200734 !access_ok(VERIFY_WRITE,
735 (void __user *)(unsigned long)mem->userspace_addr,
736 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600737 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800738 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800739 goto out;
740 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
741 goto out;
742
Xiao Guangrong28a37542011-11-24 19:04:35 +0800743 memslot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800744 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
745 npages = mem->memory_size >> PAGE_SHIFT;
746
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900747 r = -EINVAL;
748 if (npages > KVM_MEM_MAX_NR_PAGES)
749 goto out;
750
Avi Kivity6aa8b732006-12-10 02:21:36 -0800751 if (!npages)
752 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
753
Avi Kivity6aa8b732006-12-10 02:21:36 -0800754 new = old = *memslot;
755
Avi Kivitye36d96f2010-06-21 10:56:36 +0300756 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800757 new.base_gfn = base_gfn;
758 new.npages = npages;
759 new.flags = mem->flags;
760
761 /* Disallow changing a memory slot's size. */
762 r = -EINVAL;
763 if (npages && old.npages && npages != old.npages)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800764 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800765
766 /* Check for overlaps */
767 r = -EEXIST;
Alex Williamson54193692012-11-29 14:07:59 -0700768 kvm_for_each_memslot(slot, kvm->memslots) {
769 if (slot->id >= KVM_MEMORY_SLOTS || slot == memslot)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800770 continue;
Alex Williamson54193692012-11-29 14:07:59 -0700771 if (!((base_gfn + npages <= slot->base_gfn) ||
772 (base_gfn >= slot->base_gfn + slot->npages)))
Sheng Yangf78e0e22007-10-29 09:40:42 +0800773 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800774 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800775
Avi Kivity6aa8b732006-12-10 02:21:36 -0800776 /* Free page dirty bitmap if unneeded */
777 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000778 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800779
780 r = -ENOMEM;
781
782 /* Allocate if a slot is being created */
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900783 if (npages && !old.npages) {
784 new.user_alloc = user_alloc;
785 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900786
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900787 if (kvm_arch_create_memslot(&new, npages))
788 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800789 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200790
Avi Kivity6aa8b732006-12-10 02:21:36 -0800791 /* Allocate page dirty bitmap if needed */
792 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900793 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800794 goto out_free;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200795 /* destroy any largepage mappings for dirty tracking */
Avi Kivity6aa8b732006-12-10 02:21:36 -0800796 }
797
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300798 if (!npages || base_gfn != old.base_gfn) {
Xiao Guangrong28a37542011-11-24 19:04:35 +0800799 struct kvm_memory_slot *slot;
800
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200801 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100802 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
803 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200804 if (!slots)
805 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800806 slot = id_to_memslot(slots, mem->slot);
807 slot->flags |= KVM_MEMSLOT_INVALID;
808
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800809 update_memslots(slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200810
811 old_memslots = kvm->memslots;
812 rcu_assign_pointer(kvm->memslots, slots);
813 synchronize_srcu_expedited(&kvm->srcu);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300814 /* From this point no new shadow pages pointing to a deleted,
815 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200816 *
817 * validation of sp->gfn happens in:
818 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
819 * - kvm_is_visible_gfn (mmu_check_roots)
820 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300821 kvm_arch_flush_shadow_memslot(kvm, slot);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200822 kfree(old_memslots);
823 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300824
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200825 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
826 if (r)
827 goto out_free;
828
Alex Williamson32f6daa2012-04-11 09:51:49 -0600829 /* map/unmap the pages in iommu page table */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200830 if (npages) {
831 r = kvm_iommu_map_pages(kvm, &new);
832 if (r)
833 goto out_free;
Alex Williamson32f6daa2012-04-11 09:51:49 -0600834 } else
835 kvm_iommu_unmap_pages(kvm, &old);
Andrea Arcangeli604b38a2008-07-25 16:32:03 +0200836
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200837 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100838 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
839 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200840 if (!slots)
841 goto out_free;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200842
843 /* actual memory is freed via old in kvm_free_physmem_slot below */
844 if (!npages) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200845 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900846 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200847 }
848
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800849 update_memslots(slots, &new);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200850 old_memslots = kvm->memslots;
851 rcu_assign_pointer(kvm->memslots, slots);
852 synchronize_srcu_expedited(&kvm->srcu);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800853
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200854 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800855
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200856 kvm_free_physmem_slot(&old, &new);
857 kfree(old_memslots);
858
Avi Kivity6aa8b732006-12-10 02:21:36 -0800859 return 0;
860
Sheng Yangf78e0e22007-10-29 09:40:42 +0800861out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800862 kvm_free_physmem_slot(&new, &old);
863out:
864 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200865
866}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800867EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
868
869int kvm_set_memory_region(struct kvm *kvm,
870 struct kvm_userspace_memory_region *mem,
871 int user_alloc)
872{
873 int r;
874
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200875 mutex_lock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800876 r = __kvm_set_memory_region(kvm, mem, user_alloc);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200877 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800878 return r;
879}
Izik Eidus210c7c42007-10-24 23:52:57 +0200880EXPORT_SYMBOL_GPL(kvm_set_memory_region);
881
Carsten Otte1fe779f2007-10-29 16:08:35 +0100882int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
883 struct
884 kvm_userspace_memory_region *mem,
885 int user_alloc)
Izik Eidus210c7c42007-10-24 23:52:57 +0200886{
Izik Eiduse0d62c72007-10-24 23:57:46 +0200887 if (mem->slot >= KVM_MEMORY_SLOTS)
888 return -EINVAL;
Izik Eidus210c7c42007-10-24 23:52:57 +0200889 return kvm_set_memory_region(kvm, mem, user_alloc);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800890}
891
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800892int kvm_get_dirty_log(struct kvm *kvm,
893 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800894{
895 struct kvm_memory_slot *memslot;
896 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900897 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800898 unsigned long any = 0;
899
Avi Kivity6aa8b732006-12-10 02:21:36 -0800900 r = -EINVAL;
901 if (log->slot >= KVM_MEMORY_SLOTS)
902 goto out;
903
Xiao Guangrong28a37542011-11-24 19:04:35 +0800904 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800905 r = -ENOENT;
906 if (!memslot->dirty_bitmap)
907 goto out;
908
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900909 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800910
Uri Lublincd1a4a92007-02-22 16:43:09 +0200911 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800912 any = memslot->dirty_bitmap[i];
913
914 r = -EFAULT;
915 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
916 goto out;
917
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800918 if (any)
919 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800920
921 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800922out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800923 return r;
924}
925
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900926bool kvm_largepages_enabled(void)
927{
928 return largepages_enabled;
929}
930
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300931void kvm_disable_largepages(void)
932{
933 largepages_enabled = false;
934}
935EXPORT_SYMBOL_GPL(kvm_disable_largepages);
936
Gleb Natapov49c77542010-10-18 15:22:23 +0200937struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
938{
939 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
940}
Avi Kivitya1f4d3952010-06-21 11:44:20 +0300941EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800942
Izik Eiduse0d62c72007-10-24 23:57:46 +0200943int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
944{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800945 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +0200946
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800947 if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
948 memslot->flags & KVM_MEMSLOT_INVALID)
949 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200950
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800951 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +0200952}
953EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
954
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +0100955unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
956{
957 struct vm_area_struct *vma;
958 unsigned long addr, size;
959
960 size = PAGE_SIZE;
961
962 addr = gfn_to_hva(kvm, gfn);
963 if (kvm_is_error_hva(addr))
964 return PAGE_SIZE;
965
966 down_read(&current->mm->mmap_sem);
967 vma = find_vma(current->mm, addr);
968 if (!vma)
969 goto out;
970
971 size = vma_kernel_pagesize(vma);
972
973out:
974 up_read(&current->mm->mmap_sem);
975
976 return size;
977}
978
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800979static bool memslot_is_readonly(struct kvm_memory_slot *slot)
980{
981 return slot->flags & KVM_MEM_READONLY;
982}
983
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800984static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
985 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +0200986{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200987 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +0800988 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +0800989
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800990 if (memslot_is_readonly(slot) && write)
991 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +0800992
993 if (nr_pages)
994 *nr_pages = slot->npages - (gfn - slot->base_gfn);
995
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800996 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +0200997}
Xiao Guangrong48987782010-08-22 19:11:43 +0800998
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800999static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1000 gfn_t *nr_pages)
1001{
1002 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1003}
1004
1005unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1006 gfn_t gfn)
1007{
1008 return gfn_to_hva_many(slot, gfn, NULL);
1009}
1010EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1011
Xiao Guangrong48987782010-08-22 19:11:43 +08001012unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1013{
Gleb Natapov49c77542010-10-18 15:22:23 +02001014 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001015}
Sheng Yang0d150292008-04-25 21:44:50 +08001016EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001017
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001018/*
1019 * The hva returned by this function is only allowed to be read.
1020 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1021 */
1022static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001023{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001024 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001025}
1026
1027static int kvm_read_hva(void *data, void __user *hva, int len)
1028{
1029 return __copy_from_user(data, hva, len);
1030}
1031
1032static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1033{
1034 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001035}
1036
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001037int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1038 unsigned long start, int write, struct page **page)
1039{
1040 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1041
1042 if (write)
1043 flags |= FOLL_WRITE;
1044
1045 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1046}
1047
Huang Yingfafc3db2011-01-30 11:15:49 +08001048static inline int check_user_page_hwpoison(unsigned long addr)
1049{
1050 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1051
1052 rc = __get_user_pages(current, current->mm, addr, 1,
1053 flags, NULL, NULL, NULL);
1054 return rc == -EHWPOISON;
1055}
1056
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001057/*
1058 * The atomic path to get the writable pfn which will be stored in @pfn,
1059 * true indicates success, otherwise false is returned.
1060 */
1061static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1062 bool write_fault, bool *writable, pfn_t *pfn)
1063{
1064 struct page *page[1];
1065 int npages;
1066
1067 if (!(async || atomic))
1068 return false;
1069
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001070 /*
1071 * Fast pin a writable pfn only if it is a write fault request
1072 * or the caller allows to map a writable pfn for a read fault
1073 * request.
1074 */
1075 if (!(write_fault || writable))
1076 return false;
1077
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001078 npages = __get_user_pages_fast(addr, 1, 1, page);
1079 if (npages == 1) {
1080 *pfn = page_to_pfn(page[0]);
1081
1082 if (writable)
1083 *writable = true;
1084 return true;
1085 }
1086
1087 return false;
1088}
1089
1090/*
1091 * The slow path to get the pfn of the specified host virtual address,
1092 * 1 indicates success, -errno is returned if error is detected.
1093 */
1094static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1095 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001096{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001097 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001098 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001099
1100 might_sleep();
1101
1102 if (writable)
1103 *writable = write_fault;
1104
1105 if (async) {
1106 down_read(&current->mm->mmap_sem);
1107 npages = get_user_page_nowait(current, current->mm,
1108 addr, write_fault, page);
1109 up_read(&current->mm->mmap_sem);
1110 } else
1111 npages = get_user_pages_fast(addr, 1, write_fault,
1112 page);
1113 if (npages != 1)
1114 return npages;
1115
1116 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001117 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001118 struct page *wpage[1];
1119
1120 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1121 if (npages == 1) {
1122 *writable = true;
1123 put_page(page[0]);
1124 page[0] = wpage[0];
1125 }
1126
1127 npages = 1;
1128 }
1129 *pfn = page_to_pfn(page[0]);
1130 return npages;
1131}
1132
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001133static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1134{
1135 if (unlikely(!(vma->vm_flags & VM_READ)))
1136 return false;
1137
1138 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1139 return false;
1140
1141 return true;
1142}
1143
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001144/*
1145 * Pin guest page in memory and return its pfn.
1146 * @addr: host virtual address which maps memory to the guest
1147 * @atomic: whether this function can sleep
1148 * @async: whether this function need to wait IO complete if the
1149 * host page is not in the memory
1150 * @write_fault: whether we should get a writable host page
1151 * @writable: whether it allows to map a writable host page for !@write_fault
1152 *
1153 * The function will map a writable host page for these two cases:
1154 * 1): @write_fault = true
1155 * 2): @write_fault = false && @writable, @writable will tell the caller
1156 * whether the mapping is writable.
1157 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001158static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1159 bool write_fault, bool *writable)
1160{
1161 struct vm_area_struct *vma;
1162 pfn_t pfn = 0;
1163 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001164
Gleb Natapovaf585b92010-10-14 11:22:46 +02001165 /* we can do it either atomically or asynchronously, not both */
1166 BUG_ON(atomic && async);
1167
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001168 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1169 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001170
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001171 if (atomic)
1172 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001173
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001174 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1175 if (npages == 1)
1176 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001177
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001178 down_read(&current->mm->mmap_sem);
1179 if (npages == -EHWPOISON ||
1180 (!async && check_user_page_hwpoison(addr))) {
1181 pfn = KVM_PFN_ERR_HWPOISON;
1182 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001183 }
Izik Eidus539cb662007-11-11 22:05:04 +02001184
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001185 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001186
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001187 if (vma == NULL)
1188 pfn = KVM_PFN_ERR_FAULT;
1189 else if ((vma->vm_flags & VM_PFNMAP)) {
1190 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1191 vma->vm_pgoff;
1192 BUG_ON(!kvm_is_mmio_pfn(pfn));
1193 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001194 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001195 *async = true;
1196 pfn = KVM_PFN_ERR_FAULT;
1197 }
1198exit:
1199 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001200 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001201}
1202
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001203static pfn_t
1204__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1205 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001206{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001207 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1208
1209 if (addr == KVM_HVA_ERR_RO_BAD)
1210 return KVM_PFN_ERR_RO_FAULT;
1211
1212 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001213 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001214
1215 /* Do not map writable pfn in the readonly memslot. */
1216 if (writable && memslot_is_readonly(slot)) {
1217 *writable = false;
1218 writable = NULL;
1219 }
1220
1221 return hva_to_pfn(addr, atomic, async, write_fault,
1222 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001223}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001224
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001225static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1226 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001227{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001228 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001229
Gleb Natapovaf585b92010-10-14 11:22:46 +02001230 if (async)
1231 *async = false;
1232
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001233 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001234
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001235 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1236 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001237}
1238
1239pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1240{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001241 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001242}
1243EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1244
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001245pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1246 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001247{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001248 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001249}
1250EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1251
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001252pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1253{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001254 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001255}
Anthony Liguori35149e22008-04-02 14:46:56 -05001256EXPORT_SYMBOL_GPL(gfn_to_pfn);
1257
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001258pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1259 bool *writable)
1260{
1261 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1262}
1263EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1264
Xiao Guangrongd5661042012-07-17 21:56:16 +08001265pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001266{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001267 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001268}
1269
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001270pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1271{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001272 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001273}
1274EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1275
Xiao Guangrong48987782010-08-22 19:11:43 +08001276int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1277 int nr_pages)
1278{
1279 unsigned long addr;
1280 gfn_t entry;
1281
Gleb Natapov49c77542010-10-18 15:22:23 +02001282 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001283 if (kvm_is_error_hva(addr))
1284 return -1;
1285
1286 if (entry < nr_pages)
1287 return 0;
1288
1289 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1290}
1291EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1292
Xiao Guangronga2766322012-07-26 11:58:59 +08001293static struct page *kvm_pfn_to_page(pfn_t pfn)
1294{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001295 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001296 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001297
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001298 if (kvm_is_mmio_pfn(pfn)) {
1299 WARN_ON(1);
1300 return KVM_ERR_PTR_BAD_PAGE;
1301 }
1302
Xiao Guangronga2766322012-07-26 11:58:59 +08001303 return pfn_to_page(pfn);
1304}
1305
Anthony Liguori35149e22008-04-02 14:46:56 -05001306struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1307{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001308 pfn_t pfn;
1309
1310 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001311
Xiao Guangronga2766322012-07-26 11:58:59 +08001312 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001313}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001314
Avi Kivity954bbbc2007-03-30 14:02:32 +03001315EXPORT_SYMBOL_GPL(gfn_to_page);
1316
Izik Eidusb4231d62007-11-20 11:49:33 +02001317void kvm_release_page_clean(struct page *page)
1318{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001319 WARN_ON(is_error_page(page));
1320
Anthony Liguori35149e22008-04-02 14:46:56 -05001321 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001322}
1323EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1324
Anthony Liguori35149e22008-04-02 14:46:56 -05001325void kvm_release_pfn_clean(pfn_t pfn)
1326{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001327 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001328 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001329}
1330EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1331
Izik Eidusb4231d62007-11-20 11:49:33 +02001332void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001333{
Xiao Guangronga2766322012-07-26 11:58:59 +08001334 WARN_ON(is_error_page(page));
1335
Anthony Liguori35149e22008-04-02 14:46:56 -05001336 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001337}
Izik Eidusb4231d62007-11-20 11:49:33 +02001338EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001339
Anthony Liguori35149e22008-04-02 14:46:56 -05001340void kvm_release_pfn_dirty(pfn_t pfn)
1341{
1342 kvm_set_pfn_dirty(pfn);
1343 kvm_release_pfn_clean(pfn);
1344}
1345EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1346
1347void kvm_set_page_dirty(struct page *page)
1348{
1349 kvm_set_pfn_dirty(page_to_pfn(page));
1350}
1351EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1352
1353void kvm_set_pfn_dirty(pfn_t pfn)
1354{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001355 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001356 struct page *page = pfn_to_page(pfn);
1357 if (!PageReserved(page))
1358 SetPageDirty(page);
1359 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001360}
1361EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1362
1363void kvm_set_pfn_accessed(pfn_t pfn)
1364{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001365 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001366 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001367}
1368EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1369
1370void kvm_get_pfn(pfn_t pfn)
1371{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001372 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001373 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001374}
1375EXPORT_SYMBOL_GPL(kvm_get_pfn);
1376
Izik Eidus195aefd2007-10-01 22:14:18 +02001377static int next_segment(unsigned long len, int offset)
1378{
1379 if (len > PAGE_SIZE - offset)
1380 return PAGE_SIZE - offset;
1381 else
1382 return len;
1383}
1384
1385int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1386 int len)
1387{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001388 int r;
1389 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001390
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001391 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001392 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001393 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001394 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001395 if (r)
1396 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001397 return 0;
1398}
1399EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1400
1401int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1402{
1403 gfn_t gfn = gpa >> PAGE_SHIFT;
1404 int seg;
1405 int offset = offset_in_page(gpa);
1406 int ret;
1407
1408 while ((seg = next_segment(len, offset)) != 0) {
1409 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1410 if (ret < 0)
1411 return ret;
1412 offset = 0;
1413 len -= seg;
1414 data += seg;
1415 ++gfn;
1416 }
1417 return 0;
1418}
1419EXPORT_SYMBOL_GPL(kvm_read_guest);
1420
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001421int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1422 unsigned long len)
1423{
1424 int r;
1425 unsigned long addr;
1426 gfn_t gfn = gpa >> PAGE_SHIFT;
1427 int offset = offset_in_page(gpa);
1428
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001429 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001430 if (kvm_is_error_hva(addr))
1431 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001432 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001433 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001434 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001435 if (r)
1436 return -EFAULT;
1437 return 0;
1438}
1439EXPORT_SYMBOL(kvm_read_guest_atomic);
1440
Izik Eidus195aefd2007-10-01 22:14:18 +02001441int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1442 int offset, int len)
1443{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001444 int r;
1445 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001446
Izik Eiduse0506bc2007-11-11 22:10:22 +02001447 addr = gfn_to_hva(kvm, gfn);
1448 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001449 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001450 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001451 if (r)
1452 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001453 mark_page_dirty(kvm, gfn);
1454 return 0;
1455}
1456EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1457
1458int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1459 unsigned long len)
1460{
1461 gfn_t gfn = gpa >> PAGE_SHIFT;
1462 int seg;
1463 int offset = offset_in_page(gpa);
1464 int ret;
1465
1466 while ((seg = next_segment(len, offset)) != 0) {
1467 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1468 if (ret < 0)
1469 return ret;
1470 offset = 0;
1471 len -= seg;
1472 data += seg;
1473 ++gfn;
1474 }
1475 return 0;
1476}
1477
Gleb Natapov49c77542010-10-18 15:22:23 +02001478int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1479 gpa_t gpa)
1480{
1481 struct kvm_memslots *slots = kvm_memslots(kvm);
1482 int offset = offset_in_page(gpa);
1483 gfn_t gfn = gpa >> PAGE_SHIFT;
1484
1485 ghc->gpa = gpa;
1486 ghc->generation = slots->generation;
Paul Mackerras9d4cba72012-01-12 20:09:51 +00001487 ghc->memslot = gfn_to_memslot(kvm, gfn);
Gleb Natapov49c77542010-10-18 15:22:23 +02001488 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1489 if (!kvm_is_error_hva(ghc->hva))
1490 ghc->hva += offset;
1491 else
1492 return -EFAULT;
1493
1494 return 0;
1495}
1496EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1497
1498int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1499 void *data, unsigned long len)
1500{
1501 struct kvm_memslots *slots = kvm_memslots(kvm);
1502 int r;
1503
1504 if (slots->generation != ghc->generation)
1505 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1506
1507 if (kvm_is_error_hva(ghc->hva))
1508 return -EFAULT;
1509
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001510 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001511 if (r)
1512 return -EFAULT;
1513 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1514
1515 return 0;
1516}
1517EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1518
Gleb Natapove03b6442011-07-11 15:28:11 -04001519int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1520 void *data, unsigned long len)
1521{
1522 struct kvm_memslots *slots = kvm_memslots(kvm);
1523 int r;
1524
1525 if (slots->generation != ghc->generation)
1526 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1527
1528 if (kvm_is_error_hva(ghc->hva))
1529 return -EFAULT;
1530
1531 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1532 if (r)
1533 return -EFAULT;
1534
1535 return 0;
1536}
1537EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1538
Izik Eidus195aefd2007-10-01 22:14:18 +02001539int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1540{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001541 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1542 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001543}
1544EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1545
1546int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1547{
1548 gfn_t gfn = gpa >> PAGE_SHIFT;
1549 int seg;
1550 int offset = offset_in_page(gpa);
1551 int ret;
1552
1553 while ((seg = next_segment(len, offset)) != 0) {
1554 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1555 if (ret < 0)
1556 return ret;
1557 offset = 0;
1558 len -= seg;
1559 ++gfn;
1560 }
1561 return 0;
1562}
1563EXPORT_SYMBOL_GPL(kvm_clear_guest);
1564
Gleb Natapov49c77542010-10-18 15:22:23 +02001565void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1566 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001567{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001568 if (memslot && memslot->dirty_bitmap) {
1569 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001570
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001571 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001572 }
1573}
1574
Gleb Natapov49c77542010-10-18 15:22:23 +02001575void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1576{
1577 struct kvm_memory_slot *memslot;
1578
1579 memslot = gfn_to_memslot(kvm, gfn);
1580 mark_page_dirty_in_slot(kvm, memslot, gfn);
1581}
1582
Eddie Dongb6958ce2007-07-18 12:15:21 +03001583/*
1584 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1585 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001586void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001587{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001588 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001589
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001590 for (;;) {
1591 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001592
Gleb Natapova1b37102009-07-09 15:33:52 +03001593 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001594 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001595 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001596 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001597 if (kvm_cpu_has_pending_timer(vcpu))
1598 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001599 if (signal_pending(current))
1600 break;
1601
Eddie Dongb6958ce2007-07-18 12:15:21 +03001602 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001603 }
1604
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001605 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001606}
1607
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001608#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001609/*
1610 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1611 */
1612void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1613{
1614 int me;
1615 int cpu = vcpu->cpu;
1616 wait_queue_head_t *wqp;
1617
1618 wqp = kvm_arch_vcpu_wq(vcpu);
1619 if (waitqueue_active(wqp)) {
1620 wake_up_interruptible(wqp);
1621 ++vcpu->stat.halt_wakeup;
1622 }
1623
1624 me = get_cpu();
1625 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1626 if (kvm_arch_vcpu_should_kick(vcpu))
1627 smp_send_reschedule(cpu);
1628 put_cpu();
1629}
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001630#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001631
Avi Kivity6aa8b732006-12-10 02:21:36 -08001632void kvm_resched(struct kvm_vcpu *vcpu)
1633{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001634 if (!need_resched())
1635 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001636 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001637}
1638EXPORT_SYMBOL_GPL(kvm_resched);
1639
Konstantin Weitz41628d32012-04-25 15:30:38 +02001640bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1641{
1642 struct pid *pid;
1643 struct task_struct *task = NULL;
1644
1645 rcu_read_lock();
1646 pid = rcu_dereference(target->pid);
1647 if (pid)
1648 task = get_pid_task(target->pid, PIDTYPE_PID);
1649 rcu_read_unlock();
1650 if (!task)
1651 return false;
1652 if (task->flags & PF_VCPU) {
1653 put_task_struct(task);
1654 return false;
1655 }
1656 if (yield_to(task, 1)) {
1657 put_task_struct(task);
1658 return true;
1659 }
1660 put_task_struct(task);
1661 return false;
1662}
1663EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1664
Raghavendra K T06e48c52012-07-19 15:17:52 +05301665#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1666/*
1667 * Helper that checks whether a VCPU is eligible for directed yield.
1668 * Most eligible candidate to yield is decided by following heuristics:
1669 *
1670 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1671 * (preempted lock holder), indicated by @in_spin_loop.
1672 * Set at the beiginning and cleared at the end of interception/PLE handler.
1673 *
1674 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1675 * chance last time (mostly it has become eligible now since we have probably
1676 * yielded to lockholder in last iteration. This is done by toggling
1677 * @dy_eligible each time a VCPU checked for eligibility.)
1678 *
1679 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1680 * to preempted lock-holder could result in wrong VCPU selection and CPU
1681 * burning. Giving priority for a potential lock-holder increases lock
1682 * progress.
1683 *
1684 * Since algorithm is based on heuristics, accessing another VCPU data without
1685 * locking does not harm. It may result in trying to yield to same VCPU, fail
1686 * and continue with next VCPU and so on.
1687 */
1688bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1689{
1690 bool eligible;
1691
1692 eligible = !vcpu->spin_loop.in_spin_loop ||
1693 (vcpu->spin_loop.in_spin_loop &&
1694 vcpu->spin_loop.dy_eligible);
1695
1696 if (vcpu->spin_loop.in_spin_loop)
1697 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1698
1699 return eligible;
1700}
1701#endif
Rik van Riel217ece62011-02-01 09:53:28 -05001702void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001703{
Rik van Riel217ece62011-02-01 09:53:28 -05001704 struct kvm *kvm = me->kvm;
1705 struct kvm_vcpu *vcpu;
1706 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1707 int yielded = 0;
1708 int pass;
1709 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001710
Raghavendra K T4c088492012-07-18 19:07:46 +05301711 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001712 /*
1713 * We boost the priority of a VCPU that is runnable but not
1714 * currently running, because it got preempted by something
1715 * else and called schedule in __vcpu_run. Hopefully that
1716 * VCPU is holding the lock that we need and will release it.
1717 * We approximate round-robin by starting at the last boosted VCPU.
1718 */
1719 for (pass = 0; pass < 2 && !yielded; pass++) {
1720 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001721 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001722 i = last_boosted_vcpu;
1723 continue;
1724 } else if (pass && i > last_boosted_vcpu)
1725 break;
1726 if (vcpu == me)
1727 continue;
1728 if (waitqueue_active(&vcpu->wq))
1729 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301730 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1731 continue;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001732 if (kvm_vcpu_yield_to(vcpu)) {
Rik van Riel217ece62011-02-01 09:53:28 -05001733 kvm->last_boosted_vcpu = i;
1734 yielded = 1;
1735 break;
1736 }
Rik van Riel217ece62011-02-01 09:53:28 -05001737 }
1738 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301739 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301740
1741 /* Ensure vcpu is not eligible during next spinloop */
1742 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001743}
1744EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1745
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001746static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001747{
1748 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001749 struct page *page;
1750
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001751 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001752 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001753#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001754 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001755 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001756#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001757#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1758 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1759 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1760#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001761 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001762 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001763 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001764 vmf->page = page;
1765 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001766}
1767
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001768static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001769 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001770};
1771
1772static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1773{
1774 vma->vm_ops = &kvm_vcpu_vm_ops;
1775 return 0;
1776}
1777
Avi Kivitybccf2152007-02-21 18:04:26 +02001778static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1779{
1780 struct kvm_vcpu *vcpu = filp->private_data;
1781
Al Viro66c0b392008-04-19 20:33:56 +01001782 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001783 return 0;
1784}
1785
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001786static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001787 .release = kvm_vcpu_release,
1788 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001789#ifdef CONFIG_COMPAT
1790 .compat_ioctl = kvm_vcpu_compat_ioctl,
1791#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001792 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001793 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001794};
1795
1796/*
1797 * Allocates an inode for the vcpu.
1798 */
1799static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1800{
Roland Dreier628ff7c2009-12-18 09:41:24 -08001801 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
Avi Kivitybccf2152007-02-21 18:04:26 +02001802}
1803
Avi Kivityc5ea7662007-02-20 18:41:05 +02001804/*
1805 * Creates some virtual cpus. Good luck creating more than one.
1806 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001807static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001808{
1809 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001810 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001811
Gleb Natapov73880c82009-06-09 15:56:28 +03001812 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001813 if (IS_ERR(vcpu))
1814 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001815
Avi Kivity15ad7142007-07-11 18:17:21 +03001816 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1817
Avi Kivity26e52152007-11-20 15:30:24 +02001818 r = kvm_arch_vcpu_setup(vcpu);
1819 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001820 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001821
Shaohua Li11ec2802007-07-23 14:51:37 +08001822 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001823 if (!kvm_vcpu_compatible(vcpu)) {
1824 r = -EINVAL;
1825 goto unlock_vcpu_destroy;
1826 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001827 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1828 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001829 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001830 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001831
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001832 kvm_for_each_vcpu(r, v, kvm)
1833 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001834 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001835 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001836 }
1837
1838 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001839
1840 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001841 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001842 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001843 if (r < 0) {
1844 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001845 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001846 }
1847
1848 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1849 smp_wmb();
1850 atomic_inc(&kvm->online_vcpus);
1851
Gleb Natapov73880c82009-06-09 15:56:28 +03001852 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001853 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001854 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001855
Jan Kiszkad7805922011-05-23 10:33:05 +02001856unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001857 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001858vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001859 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001860 return r;
1861}
1862
Avi Kivity1961d272007-03-05 19:46:05 +02001863static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1864{
1865 if (sigset) {
1866 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1867 vcpu->sigset_active = 1;
1868 vcpu->sigset = *sigset;
1869 } else
1870 vcpu->sigset_active = 0;
1871 return 0;
1872}
1873
Avi Kivitybccf2152007-02-21 18:04:26 +02001874static long kvm_vcpu_ioctl(struct file *filp,
1875 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001876{
Avi Kivitybccf2152007-02-21 18:04:26 +02001877 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001878 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001879 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001880 struct kvm_fpu *fpu = NULL;
1881 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001882
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001883 if (vcpu->kvm->mm != current->mm)
1884 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001885
1886#if defined(CONFIG_S390) || defined(CONFIG_PPC)
1887 /*
1888 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1889 * so vcpu_load() would break it.
1890 */
1891 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1892 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1893#endif
1894
1895
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001896 r = vcpu_load(vcpu);
1897 if (r)
1898 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001899 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001900 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001901 r = -EINVAL;
1902 if (arg)
1903 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001904 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001905 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001906 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001907 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001908 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001909
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001910 r = -ENOMEM;
1911 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1912 if (!kvm_regs)
1913 goto out;
1914 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001915 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001916 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001917 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001918 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1919 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001920 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001921out_free1:
1922 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001923 break;
1924 }
1925 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001926 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001927
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001928 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001929 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1930 if (IS_ERR(kvm_regs)) {
1931 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001932 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02001933 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001934 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001935 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001936 break;
1937 }
1938 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07001939 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1940 r = -ENOMEM;
1941 if (!kvm_sregs)
1942 goto out;
1943 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001944 if (r)
1945 goto out;
1946 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001947 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001948 goto out;
1949 r = 0;
1950 break;
1951 }
1952 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02001953 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1954 if (IS_ERR(kvm_sregs)) {
1955 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08001956 kvm_sregs = NULL;
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 break;
1961 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03001962 case KVM_GET_MP_STATE: {
1963 struct kvm_mp_state mp_state;
1964
1965 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1966 if (r)
1967 goto out;
1968 r = -EFAULT;
1969 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1970 goto out;
1971 r = 0;
1972 break;
1973 }
1974 case KVM_SET_MP_STATE: {
1975 struct kvm_mp_state mp_state;
1976
1977 r = -EFAULT;
1978 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1979 goto out;
1980 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03001981 break;
1982 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001983 case KVM_TRANSLATE: {
1984 struct kvm_translation tr;
1985
1986 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00001987 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001988 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08001989 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001990 if (r)
1991 goto out;
1992 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00001993 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994 goto out;
1995 r = 0;
1996 break;
1997 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001998 case KVM_SET_GUEST_DEBUG: {
1999 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002000
2001 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002002 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002003 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002004 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002005 break;
2006 }
Avi Kivity1961d272007-03-05 19:46:05 +02002007 case KVM_SET_SIGNAL_MASK: {
2008 struct kvm_signal_mask __user *sigmask_arg = argp;
2009 struct kvm_signal_mask kvm_sigmask;
2010 sigset_t sigset, *p;
2011
2012 p = NULL;
2013 if (argp) {
2014 r = -EFAULT;
2015 if (copy_from_user(&kvm_sigmask, argp,
2016 sizeof kvm_sigmask))
2017 goto out;
2018 r = -EINVAL;
2019 if (kvm_sigmask.len != sizeof sigset)
2020 goto out;
2021 r = -EFAULT;
2022 if (copy_from_user(&sigset, sigmask_arg->sigset,
2023 sizeof sigset))
2024 goto out;
2025 p = &sigset;
2026 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002027 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002028 break;
2029 }
Avi Kivityb8836732007-04-01 16:34:31 +03002030 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002031 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2032 r = -ENOMEM;
2033 if (!fpu)
2034 goto out;
2035 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002036 if (r)
2037 goto out;
2038 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002039 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002040 goto out;
2041 r = 0;
2042 break;
2043 }
2044 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002045 fpu = memdup_user(argp, sizeof(*fpu));
2046 if (IS_ERR(fpu)) {
2047 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002048 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002049 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002050 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002051 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002052 break;
2053 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002054 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002055 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002056 }
2057out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002058 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002059 kfree(fpu);
2060 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002061 return r;
2062}
2063
Alexander Graf1dda6062011-06-08 02:45:37 +02002064#ifdef CONFIG_COMPAT
2065static long kvm_vcpu_compat_ioctl(struct file *filp,
2066 unsigned int ioctl, unsigned long arg)
2067{
2068 struct kvm_vcpu *vcpu = filp->private_data;
2069 void __user *argp = compat_ptr(arg);
2070 int r;
2071
2072 if (vcpu->kvm->mm != current->mm)
2073 return -EIO;
2074
2075 switch (ioctl) {
2076 case KVM_SET_SIGNAL_MASK: {
2077 struct kvm_signal_mask __user *sigmask_arg = argp;
2078 struct kvm_signal_mask kvm_sigmask;
2079 compat_sigset_t csigset;
2080 sigset_t sigset;
2081
2082 if (argp) {
2083 r = -EFAULT;
2084 if (copy_from_user(&kvm_sigmask, argp,
2085 sizeof kvm_sigmask))
2086 goto out;
2087 r = -EINVAL;
2088 if (kvm_sigmask.len != sizeof csigset)
2089 goto out;
2090 r = -EFAULT;
2091 if (copy_from_user(&csigset, sigmask_arg->sigset,
2092 sizeof csigset))
2093 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002094 sigset_from_compat(&sigset, &csigset);
2095 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2096 } else
2097 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002098 break;
2099 }
2100 default:
2101 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2102 }
2103
2104out:
2105 return r;
2106}
2107#endif
2108
Avi Kivitybccf2152007-02-21 18:04:26 +02002109static long kvm_vm_ioctl(struct file *filp,
2110 unsigned int ioctl, unsigned long arg)
2111{
2112 struct kvm *kvm = filp->private_data;
2113 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002114 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002115
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002116 if (kvm->mm != current->mm)
2117 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002118 switch (ioctl) {
2119 case KVM_CREATE_VCPU:
2120 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002121 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002122 case KVM_SET_USER_MEMORY_REGION: {
2123 struct kvm_userspace_memory_region kvm_userspace_mem;
2124
2125 r = -EFAULT;
2126 if (copy_from_user(&kvm_userspace_mem, argp,
2127 sizeof kvm_userspace_mem))
2128 goto out;
2129
2130 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002131 break;
2132 }
2133 case KVM_GET_DIRTY_LOG: {
2134 struct kvm_dirty_log log;
2135
2136 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002137 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002138 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002139 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002140 break;
2141 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002142#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2143 case KVM_REGISTER_COALESCED_MMIO: {
2144 struct kvm_coalesced_mmio_zone zone;
2145 r = -EFAULT;
2146 if (copy_from_user(&zone, argp, sizeof zone))
2147 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002148 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002149 break;
2150 }
2151 case KVM_UNREGISTER_COALESCED_MMIO: {
2152 struct kvm_coalesced_mmio_zone zone;
2153 r = -EFAULT;
2154 if (copy_from_user(&zone, argp, sizeof zone))
2155 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002156 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002157 break;
2158 }
2159#endif
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002160 case KVM_IRQFD: {
2161 struct kvm_irqfd data;
2162
2163 r = -EFAULT;
2164 if (copy_from_user(&data, argp, sizeof data))
2165 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002166 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecbf2009-05-20 10:30:49 -04002167 break;
2168 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002169 case KVM_IOEVENTFD: {
2170 struct kvm_ioeventfd data;
2171
2172 r = -EFAULT;
2173 if (copy_from_user(&data, argp, sizeof data))
2174 goto out;
2175 r = kvm_ioeventfd(kvm, &data);
2176 break;
2177 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002178#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2179 case KVM_SET_BOOT_CPU_ID:
2180 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002181 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002182 if (atomic_read(&kvm->online_vcpus) != 0)
2183 r = -EBUSY;
2184 else
2185 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002186 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002187 break;
2188#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002189#ifdef CONFIG_HAVE_KVM_MSI
2190 case KVM_SIGNAL_MSI: {
2191 struct kvm_msi msi;
2192
2193 r = -EFAULT;
2194 if (copy_from_user(&msi, argp, sizeof msi))
2195 goto out;
2196 r = kvm_send_userspace_msi(kvm, &msi);
2197 break;
2198 }
2199#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002200#ifdef __KVM_HAVE_IRQ_LINE
2201 case KVM_IRQ_LINE_STATUS:
2202 case KVM_IRQ_LINE: {
2203 struct kvm_irq_level irq_event;
2204
2205 r = -EFAULT;
2206 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2207 goto out;
2208
2209 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2210 if (r)
2211 goto out;
2212
2213 r = -EFAULT;
2214 if (ioctl == KVM_IRQ_LINE_STATUS) {
2215 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2216 goto out;
2217 }
2218
2219 r = 0;
2220 break;
2221 }
2222#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002223 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002224 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002225 if (r == -ENOTTY)
2226 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002227 }
2228out:
2229 return r;
2230}
2231
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002232#ifdef CONFIG_COMPAT
2233struct compat_kvm_dirty_log {
2234 __u32 slot;
2235 __u32 padding1;
2236 union {
2237 compat_uptr_t dirty_bitmap; /* one bit per page */
2238 __u64 padding2;
2239 };
2240};
2241
2242static long kvm_vm_compat_ioctl(struct file *filp,
2243 unsigned int ioctl, unsigned long arg)
2244{
2245 struct kvm *kvm = filp->private_data;
2246 int r;
2247
2248 if (kvm->mm != current->mm)
2249 return -EIO;
2250 switch (ioctl) {
2251 case KVM_GET_DIRTY_LOG: {
2252 struct compat_kvm_dirty_log compat_log;
2253 struct kvm_dirty_log log;
2254
2255 r = -EFAULT;
2256 if (copy_from_user(&compat_log, (void __user *)arg,
2257 sizeof(compat_log)))
2258 goto out;
2259 log.slot = compat_log.slot;
2260 log.padding1 = compat_log.padding1;
2261 log.padding2 = compat_log.padding2;
2262 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2263
2264 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002265 break;
2266 }
2267 default:
2268 r = kvm_vm_ioctl(filp, ioctl, arg);
2269 }
2270
2271out:
2272 return r;
2273}
2274#endif
2275
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002276static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002277{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002278 struct page *page[1];
2279 unsigned long addr;
2280 int npages;
2281 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002282 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002283
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002284 addr = gfn_to_hva(kvm, gfn);
2285 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002286 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002287
2288 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2289 NULL);
2290 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002291 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002292
2293 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002294 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002295}
2296
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002297static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002298 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002299};
2300
2301static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2302{
2303 vma->vm_ops = &kvm_vm_vm_ops;
2304 return 0;
2305}
2306
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002307static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002308 .release = kvm_vm_release,
2309 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002310#ifdef CONFIG_COMPAT
2311 .compat_ioctl = kvm_vm_compat_ioctl,
2312#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002313 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002314 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002315};
2316
Carsten Ottee08b9632012-01-04 10:25:20 +01002317static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002318{
Heiko Carstensaac87632010-10-27 17:22:10 +02002319 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002320 struct kvm *kvm;
2321
Carsten Ottee08b9632012-01-04 10:25:20 +01002322 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002323 if (IS_ERR(kvm))
2324 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002325#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2326 r = kvm_coalesced_mmio_init(kvm);
2327 if (r < 0) {
2328 kvm_put_kvm(kvm);
2329 return r;
2330 }
2331#endif
Heiko Carstensaac87632010-10-27 17:22:10 +02002332 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2333 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002334 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002335
Heiko Carstensaac87632010-10-27 17:22:10 +02002336 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002337}
2338
Avi Kivity1a811b62008-12-08 18:25:27 +02002339static long kvm_dev_ioctl_check_extension_generic(long arg)
2340{
2341 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002342 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002343 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002344 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002345#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2346 case KVM_CAP_SET_BOOT_CPU_ID:
2347#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002348 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002349#ifdef CONFIG_HAVE_KVM_MSI
2350 case KVM_CAP_SIGNAL_MSI:
2351#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002352 return 1;
Marc Zyngier9900b4b2012-06-15 15:07:02 -04002353#ifdef KVM_CAP_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002354 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002355 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002356#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002357 default:
2358 break;
2359 }
2360 return kvm_dev_ioctl_check_extension(arg);
2361}
2362
Avi Kivityf17abe92007-02-21 19:28:04 +02002363static long kvm_dev_ioctl(struct file *filp,
2364 unsigned int ioctl, unsigned long arg)
2365{
Avi Kivity07c45a32007-03-07 13:05:38 +02002366 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002367
2368 switch (ioctl) {
2369 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002370 r = -EINVAL;
2371 if (arg)
2372 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002373 r = KVM_API_VERSION;
2374 break;
2375 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002376 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002377 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002378 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002379 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002380 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002381 case KVM_GET_VCPU_MMAP_SIZE:
2382 r = -EINVAL;
2383 if (arg)
2384 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002385 r = PAGE_SIZE; /* struct kvm_run */
2386#ifdef CONFIG_X86
2387 r += PAGE_SIZE; /* pio data page */
2388#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002389#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2390 r += PAGE_SIZE; /* coalesced mmio ring page */
2391#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002392 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002393 case KVM_TRACE_ENABLE:
2394 case KVM_TRACE_PAUSE:
2395 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002396 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002397 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002398 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002399 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002400 }
2401out:
2402 return r;
2403}
2404
Avi Kivity6aa8b732006-12-10 02:21:36 -08002405static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002406 .unlocked_ioctl = kvm_dev_ioctl,
2407 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002408 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002409};
2410
2411static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002412 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002413 "kvm",
2414 &kvm_chardev_ops,
2415};
2416
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002417static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002418{
2419 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002420 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002421
Rusty Russell7f59f492008-12-07 21:25:45 +10302422 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002423 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002424
Rusty Russell7f59f492008-12-07 21:25:45 +10302425 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002426
2427 r = kvm_arch_hardware_enable(NULL);
2428
2429 if (r) {
2430 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2431 atomic_inc(&hardware_enable_failed);
2432 printk(KERN_INFO "kvm: enabling virtualization on "
2433 "CPU%d failed\n", cpu);
2434 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002435}
2436
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002437static void hardware_enable(void *junk)
2438{
Jan Kiszkae935b832011-02-08 12:55:33 +01002439 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002440 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002441 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002442}
2443
2444static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002445{
2446 int cpu = raw_smp_processor_id();
2447
Rusty Russell7f59f492008-12-07 21:25:45 +10302448 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002449 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302450 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002451 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002452}
2453
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002454static void hardware_disable(void *junk)
2455{
Jan Kiszkae935b832011-02-08 12:55:33 +01002456 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002457 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002458 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002459}
2460
Alexander Graf10474ae2009-09-15 11:37:46 +02002461static void hardware_disable_all_nolock(void)
2462{
2463 BUG_ON(!kvm_usage_count);
2464
2465 kvm_usage_count--;
2466 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002467 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002468}
2469
2470static void hardware_disable_all(void)
2471{
Jan Kiszkae935b832011-02-08 12:55:33 +01002472 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002473 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002474 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002475}
2476
2477static int hardware_enable_all(void)
2478{
2479 int r = 0;
2480
Jan Kiszkae935b832011-02-08 12:55:33 +01002481 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002482
2483 kvm_usage_count++;
2484 if (kvm_usage_count == 1) {
2485 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002486 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002487
2488 if (atomic_read(&hardware_enable_failed)) {
2489 hardware_disable_all_nolock();
2490 r = -EBUSY;
2491 }
2492 }
2493
Jan Kiszkae935b832011-02-08 12:55:33 +01002494 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002495
2496 return r;
2497}
2498
Avi Kivity774c47f2007-02-12 00:54:47 -08002499static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2500 void *v)
2501{
2502 int cpu = (long)v;
2503
Alexander Graf10474ae2009-09-15 11:37:46 +02002504 if (!kvm_usage_count)
2505 return NOTIFY_OK;
2506
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002507 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002508 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002509 case CPU_DYING:
Avi Kivity6ec8a852007-08-19 15:57:26 +03002510 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2511 cpu);
2512 hardware_disable(NULL);
2513 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002514 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002515 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2516 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002517 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002518 break;
2519 }
2520 return NOTIFY_OK;
2521}
2522
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002523
Avi Kivityb7c41452010-12-02 17:52:50 +02002524asmlinkage void kvm_spurious_fault(void)
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002525{
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002526 /* Fault while not rebooting. We want the trace. */
2527 BUG();
2528}
Avi Kivityb7c41452010-12-02 17:52:50 +02002529EXPORT_SYMBOL_GPL(kvm_spurious_fault);
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002530
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002531static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002532 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002533{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002534 /*
2535 * Some (well, at least mine) BIOSes hang on reboot if
2536 * in vmx root mode.
2537 *
2538 * And Intel TXT required VMX off for all cpu when system shutdown.
2539 */
2540 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2541 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002542 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002543 return NOTIFY_OK;
2544}
2545
2546static struct notifier_block kvm_reboot_notifier = {
2547 .notifier_call = kvm_reboot,
2548 .priority = 0,
2549};
2550
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002551static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002552{
2553 int i;
2554
2555 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002556 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002557
2558 kvm_iodevice_destructor(pos);
2559 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002560 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002561}
2562
Sasha Levin743eeb02011-07-27 16:00:48 +03002563int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2564{
2565 const struct kvm_io_range *r1 = p1;
2566 const struct kvm_io_range *r2 = p2;
2567
2568 if (r1->addr < r2->addr)
2569 return -1;
2570 if (r1->addr + r1->len > r2->addr + r2->len)
2571 return 1;
2572 return 0;
2573}
2574
2575int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2576 gpa_t addr, int len)
2577{
Sasha Levin743eeb02011-07-27 16:00:48 +03002578 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2579 .addr = addr,
2580 .len = len,
2581 .dev = dev,
2582 };
2583
2584 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2585 kvm_io_bus_sort_cmp, NULL);
2586
2587 return 0;
2588}
2589
2590int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2591 gpa_t addr, int len)
2592{
2593 struct kvm_io_range *range, key;
2594 int off;
2595
2596 key = (struct kvm_io_range) {
2597 .addr = addr,
2598 .len = len,
2599 };
2600
2601 range = bsearch(&key, bus->range, bus->dev_count,
2602 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2603 if (range == NULL)
2604 return -ENOENT;
2605
2606 off = range - bus->range;
2607
2608 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2609 off--;
2610
2611 return off;
2612}
2613
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002614/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002615int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002616 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002617{
Sasha Levin743eeb02011-07-27 16:00:48 +03002618 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002619 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002620 struct kvm_io_range range;
2621
2622 range = (struct kvm_io_range) {
2623 .addr = addr,
2624 .len = len,
2625 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002626
2627 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002628 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2629 if (idx < 0)
2630 return -EOPNOTSUPP;
2631
2632 while (idx < bus->dev_count &&
2633 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2634 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002635 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002636 idx++;
2637 }
2638
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002639 return -EOPNOTSUPP;
2640}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002641
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002642/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002643int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2644 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002645{
Sasha Levin743eeb02011-07-27 16:00:48 +03002646 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002647 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002648 struct kvm_io_range range;
2649
2650 range = (struct kvm_io_range) {
2651 .addr = addr,
2652 .len = len,
2653 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002654
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002655 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002656 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2657 if (idx < 0)
2658 return -EOPNOTSUPP;
2659
2660 while (idx < bus->dev_count &&
2661 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2662 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002663 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002664 idx++;
2665 }
2666
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002667 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002668}
2669
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002670/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002671int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2672 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002673{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002674 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002675
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002676 bus = kvm->buses[bus_idx];
Amos Konga13007162012-03-09 12:17:32 +08002677 if (bus->dev_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002678 return -ENOSPC;
2679
Amos Konga13007162012-03-09 12:17:32 +08002680 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2681 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002682 if (!new_bus)
2683 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002684 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2685 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002686 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002687 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2688 synchronize_srcu_expedited(&kvm->srcu);
2689 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002690
2691 return 0;
2692}
2693
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002694/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002695int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2696 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002697{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002698 int i, r;
2699 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002700
Sasha Levincdfca7b2011-12-04 19:36:28 +02002701 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002702 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002703 for (i = 0; i < bus->dev_count; i++)
2704 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002705 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002706 break;
2707 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002708
Amos Konga13007162012-03-09 12:17:32 +08002709 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002710 return r;
Amos Konga13007162012-03-09 12:17:32 +08002711
2712 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2713 sizeof(struct kvm_io_range)), GFP_KERNEL);
2714 if (!new_bus)
2715 return -ENOMEM;
2716
2717 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2718 new_bus->dev_count--;
2719 memcpy(new_bus->range + i, bus->range + i + 1,
2720 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002721
2722 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2723 synchronize_srcu_expedited(&kvm->srcu);
2724 kfree(bus);
2725 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002726}
2727
Avi Kivity774c47f2007-02-12 00:54:47 -08002728static struct notifier_block kvm_cpu_notifier = {
2729 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08002730};
2731
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002732static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002733{
2734 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002735 struct kvm *kvm;
2736
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002737 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002738 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02002739 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002740 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01002741 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002742 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002743}
2744
2745DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2746
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002747static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03002748{
2749 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002750 struct kvm *kvm;
2751 struct kvm_vcpu *vcpu;
2752 int i;
2753
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002754 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002755 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03002756 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002757 kvm_for_each_vcpu(i, vcpu, kvm)
2758 *val += *(u32 *)((void *)vcpu + offset);
2759
Jan Kiszkae935b832011-02-08 12:55:33 +01002760 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002761 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002762}
2763
Avi Kivityba1389b2007-11-18 16:24:12 +02002764DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2765
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002766static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02002767 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2768 [KVM_STAT_VM] = &vm_stat_fops,
2769};
Avi Kivity1165f5f2007-04-19 17:27:43 +03002770
Hamo4f69b682011-12-15 14:23:16 +08002771static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002772{
Hamo4f69b682011-12-15 14:23:16 +08002773 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002774 struct kvm_stats_debugfs_item *p;
2775
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002776 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08002777 if (kvm_debugfs_dir == NULL)
2778 goto out;
2779
2780 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002781 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03002782 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02002783 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08002784 if (p->dentry == NULL)
2785 goto out_dir;
2786 }
2787
2788 return 0;
2789
2790out_dir:
2791 debugfs_remove_recursive(kvm_debugfs_dir);
2792out:
2793 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002794}
2795
2796static void kvm_exit_debug(void)
2797{
2798 struct kvm_stats_debugfs_item *p;
2799
2800 for (p = debugfs_entries; p->name; ++p)
2801 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002802 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002803}
2804
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002805static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002806{
Alexander Graf10474ae2009-09-15 11:37:46 +02002807 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002808 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002809 return 0;
2810}
2811
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002812static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002813{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002814 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01002815 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002816 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002817 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08002818}
2819
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002820static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08002821 .suspend = kvm_suspend,
2822 .resume = kvm_resume,
2823};
2824
Avi Kivity15ad7142007-07-11 18:17:21 +03002825static inline
2826struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2827{
2828 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2829}
2830
2831static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2832{
2833 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2834
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002835 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002836}
2837
2838static void kvm_sched_out(struct preempt_notifier *pn,
2839 struct task_struct *next)
2840{
2841 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2842
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002843 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002844}
2845
Avi Kivity0ee75be2010-04-28 15:39:01 +03002846int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10002847 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002848{
2849 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002850 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002851
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002852 r = kvm_arch_init(opaque);
2853 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002854 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002855
Avi Kivity8437a612009-06-06 14:52:35 -07002856 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10302857 r = -ENOMEM;
2858 goto out_free_0;
2859 }
2860
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002861 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002862 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10302863 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002864
Yang, Sheng002c7f72007-07-31 14:23:01 +03002865 for_each_online_cpu(cpu) {
2866 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002867 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02002868 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03002869 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002870 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002871 }
2872
Avi Kivity774c47f2007-02-12 00:54:47 -08002873 r = register_cpu_notifier(&kvm_cpu_notifier);
2874 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002875 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002876 register_reboot_notifier(&kvm_reboot_notifier);
2877
Rusty Russellc16f8622007-07-30 21:12:19 +10002878 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03002879 if (!vcpu_align)
2880 vcpu_align = __alignof__(struct kvm_vcpu);
2881 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08002882 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10002883 if (!kvm_vcpu_cache) {
2884 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002885 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10002886 }
2887
Gleb Natapovaf585b92010-10-14 11:22:46 +02002888 r = kvm_async_pf_init();
2889 if (r)
2890 goto out_free;
2891
Avi Kivity6aa8b732006-12-10 02:21:36 -08002892 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002893 kvm_vm_fops.owner = module;
2894 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002895
2896 r = misc_register(&kvm_dev);
2897 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04002898 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02002899 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002900 }
2901
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002902 register_syscore_ops(&kvm_syscore_ops);
2903
Avi Kivity15ad7142007-07-11 18:17:21 +03002904 kvm_preempt_ops.sched_in = kvm_sched_in;
2905 kvm_preempt_ops.sched_out = kvm_sched_out;
2906
Hamo4f69b682011-12-15 14:23:16 +08002907 r = kvm_init_debug();
2908 if (r) {
2909 printk(KERN_ERR "kvm: create debugfs files failed\n");
2910 goto out_undebugfs;
2911 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07002912
Avi Kivityc7addb92007-09-16 18:58:32 +02002913 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002914
Hamo4f69b682011-12-15 14:23:16 +08002915out_undebugfs:
2916 unregister_syscore_ops(&kvm_syscore_ops);
Gleb Natapovaf585b92010-10-14 11:22:46 +02002917out_unreg:
2918 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002919out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10002920 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002921out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002922 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08002923 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002924out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08002925out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002926 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10302927out_free_0a:
2928 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002929out_free_0:
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002930 kvm_arch_exit();
Zhang Xiantaod23087842007-11-29 15:35:39 +08002931out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002932 return r;
2933}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002934EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002935
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002936void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002937{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07002938 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002939 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10002940 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02002941 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002942 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002943 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002944 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002945 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002946 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002947 kvm_arch_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10302948 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002949}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002950EXPORT_SYMBOL_GPL(kvm_exit);