blob: adc68feb5c5a338080ac8ca2385f73577ac6c29a [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Hollis Blancharde2174022007-12-03 15:30:24 -060019#include "iodev.h"
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
55#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020056#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080057
Laurent Vivier5f94c172008-05-30 16:05:54 +020058#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020059#include "async_pf.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020060
Marcelo Tosatti229456f2009-06-17 09:22:14 -030061#define CREATE_TRACE_POINTS
62#include <trace/events/kvm.h>
63
Avi Kivity6aa8b732006-12-10 02:21:36 -080064MODULE_AUTHOR("Qumranet");
65MODULE_LICENSE("GPL");
66
Marcelo Tosattifa40a822009-06-04 15:08:24 -030067/*
68 * Ordering of locks:
69 *
Sheng Yangfae3a352009-12-15 10:28:07 +080070 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030071 */
72
Jan Kiszkae935b832011-02-08 12:55:33 +010073DEFINE_RAW_SPINLOCK(kvm_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080074LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080075
Rusty Russell7f59f492008-12-07 21:25:45 +103076static cpumask_var_t cpus_hardware_enabled;
Alexander Graf10474ae2009-09-15 11:37:46 +020077static int kvm_usage_count = 0;
78static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030079
Rusty Russellc16f8622007-07-30 21:12:19 +100080struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +030082
Avi Kivity15ad7142007-07-11 18:17:21 +030083static __read_mostly struct preempt_ops kvm_preempt_ops;
84
Hollis Blanchard76f7c872008-04-15 16:05:42 -050085struct dentry *kvm_debugfs_dir;
Avi Kivity6aa8b732006-12-10 02:21:36 -080086
Avi Kivitybccf2152007-02-21 18:04:26 +020087static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg);
Alexander Graf1dda6062011-06-08 02:45:37 +020089#ifdef CONFIG_COMPAT
90static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 unsigned long arg);
92#endif
Alexander Graf10474ae2009-09-15 11:37:46 +020093static int hardware_enable_all(void);
94static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +020095
Marcelo Tosattie93f8a02009-12-23 14:35:24 -020096static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
Avi Kivityb7c41452010-12-02 17:52:50 +020098bool kvm_rebooting;
99EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300100
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300101static bool largepages_enabled = true;
102
Xiao Guangronga2766322012-07-26 11:58:59 +0800103bool kvm_is_mmio_pfn(pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300104{
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100105 if (pfn_valid(pfn)) {
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800106 int reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800107 struct page *tail = pfn_to_page(pfn);
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800110 if (head != tail) {
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800111 /*
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
119 * what we just read.
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800120 */
Andrea Arcangeli22e5c472011-01-13 15:47:20 -0800121 smp_rmb();
122 if (PageTail(tail))
123 return reserved;
Andrea Arcangeli936a5fe2011-01-13 15:46:48 -0800124 }
125 return PageReserved(tail);
Joerg Roedelfc5659c2009-02-18 14:08:58 +0100126 }
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300127
128 return true;
129}
130
Avi Kivity6aa8b732006-12-10 02:21:36 -0800131/*
132 * Switches to specified vcpu, until a matching vcpu_put()
133 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300134int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800135{
Avi Kivity15ad7142007-07-11 18:17:21 +0300136 int cpu;
137
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300138 if (mutex_lock_killable(&vcpu->mutex))
139 return -EINTR;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
145 synchronize_rcu();
146 put_pid(oldpid);
147 }
Avi Kivity15ad7142007-07-11 18:17:21 +0300148 cpu = get_cpu();
149 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200150 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300151 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300152 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200153}
154
Carsten Otte313a3dc2007-10-11 19:16:52 +0200155void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800156{
Avi Kivity15ad7142007-07-11 18:17:21 +0300157 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200158 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300159 preempt_notifier_unregister(&vcpu->preempt_notifier);
160 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800161 mutex_unlock(&vcpu->mutex);
162}
163
Avi Kivityd9e368d2007-06-07 19:18:30 +0300164static void ack_flush(void *_completed)
165{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300166}
167
Rusty Russell49846892008-12-08 20:26:24 +1030168static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300169{
Avi Kivity597a5f52008-07-20 14:24:22 +0300170 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030171 cpumask_var_t cpus;
172 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300173 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300174
Li Zefan79f55992009-06-15 14:58:26 +0800175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030176
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800177 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300178 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800179 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300180 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800181
182 /* Set ->requests bit before we read ->mode */
183 smp_mb();
184
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030187 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300188 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
193 else
194 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800195 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030196 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030197 return called;
198}
199
200void kvm_flush_remote_tlbs(struct kvm *kvm)
201{
Alex Shibec87d62012-03-04 13:13:38 +0800202 long dirty_count = kvm->tlbs_dirty;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800203
204 smp_mb();
Rusty Russell49846892008-12-08 20:26:24 +1030205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300208}
209
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500210void kvm_reload_remote_mmus(struct kvm *kvm)
211{
Rusty Russell49846892008-12-08 20:26:24 +1030212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500213}
214
Marcelo Tosattid8281992012-11-27 23:29:01 -0200215void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216{
217 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218}
219
Yang Zhangc7c9c562013-01-25 10:18:51 +0800220void kvm_make_update_eoibitmap_request(struct kvm *kvm)
221{
222 make_all_cpus_request(kvm, KVM_REQ_EOIBITMAP);
223}
224
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000225int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
226{
227 struct page *page;
228 int r;
229
230 mutex_init(&vcpu->mutex);
231 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000232 vcpu->kvm = kvm;
233 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500234 vcpu->pid = NULL;
Eddie Dongb6958ce2007-07-18 12:15:21 +0300235 init_waitqueue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200236 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000237
238 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
239 if (!page) {
240 r = -ENOMEM;
241 goto fail;
242 }
243 vcpu->run = page_address(page);
244
Raghavendra K T4c088492012-07-18 19:07:46 +0530245 kvm_vcpu_set_in_spin_loop(vcpu, false);
246 kvm_vcpu_set_dy_eligible(vcpu, false);
247
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800248 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000249 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800250 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000251 return 0;
252
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000253fail_free_run:
254 free_page((unsigned long)vcpu->run);
255fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000256 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000257}
258EXPORT_SYMBOL_GPL(kvm_vcpu_init);
259
260void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
261{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500262 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800263 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000264 free_page((unsigned long)vcpu->run);
265}
266EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
267
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200268#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
269static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
270{
271 return container_of(mn, struct kvm, mmu_notifier);
272}
273
274static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
275 struct mm_struct *mm,
276 unsigned long address)
277{
278 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200279 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200280
281 /*
282 * When ->invalidate_page runs, the linux pte has been zapped
283 * already but the page is still allocated until
284 * ->invalidate_page returns. So if we increase the sequence
285 * here the kvm page fault will notice if the spte can't be
286 * established because the page is going to be freed. If
287 * instead the kvm page fault establishes the spte before
288 * ->invalidate_page runs, kvm_unmap_hva will release it
289 * before returning.
290 *
291 * The sequence increase only need to be seen at spin_unlock
292 * time, and not at spin_lock time.
293 *
294 * Increasing the sequence after the spin_unlock would be
295 * unsafe because the kvm page fault could then establish the
296 * pte after kvm_unmap_hva returned, without noticing the page
297 * is going to be freed.
298 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200299 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200300 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900301
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200302 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800303 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200304 /* we've to flush the tlb before the pages can be freed */
305 if (need_tlb_flush)
306 kvm_flush_remote_tlbs(kvm);
307
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900308 spin_unlock(&kvm->mmu_lock);
309 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200310}
311
Izik Eidus3da0dd42009-09-23 21:47:18 +0300312static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
313 struct mm_struct *mm,
314 unsigned long address,
315 pte_t pte)
316{
317 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200318 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300319
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200320 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300321 spin_lock(&kvm->mmu_lock);
322 kvm->mmu_notifier_seq++;
323 kvm_set_spte_hva(kvm, address, pte);
324 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200325 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300326}
327
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200328static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long start,
331 unsigned long end)
332{
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200334 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200335
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200336 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200337 spin_lock(&kvm->mmu_lock);
338 /*
339 * The count increase must become visible at unlock time as no
340 * spte can be established without taking the mmu_lock and
341 * count is also read inside the mmu_lock critical section.
342 */
343 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900344 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800345 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200346 /* we've to flush the tlb before the pages can be freed */
347 if (need_tlb_flush)
348 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900349
350 spin_unlock(&kvm->mmu_lock);
351 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200352}
353
354static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
355 struct mm_struct *mm,
356 unsigned long start,
357 unsigned long end)
358{
359 struct kvm *kvm = mmu_notifier_to_kvm(mn);
360
361 spin_lock(&kvm->mmu_lock);
362 /*
363 * This sequence increase will notify the kvm page fault that
364 * the page that is going to be mapped in the spte could have
365 * been freed.
366 */
367 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000368 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200369 /*
370 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000371 * below count decrease, which is ensured by the smp_wmb above
372 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200373 */
374 kvm->mmu_notifier_count--;
375 spin_unlock(&kvm->mmu_lock);
376
377 BUG_ON(kvm->mmu_notifier_count < 0);
378}
379
380static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
381 struct mm_struct *mm,
382 unsigned long address)
383{
384 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200385 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200386
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200387 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200388 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200389
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900390 young = kvm_age_hva(kvm, address);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200391 if (young)
392 kvm_flush_remote_tlbs(kvm);
393
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900394 spin_unlock(&kvm->mmu_lock);
395 srcu_read_unlock(&kvm->srcu, idx);
396
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200397 return young;
398}
399
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800400static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
401 struct mm_struct *mm,
402 unsigned long address)
403{
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
405 int young, idx;
406
407 idx = srcu_read_lock(&kvm->srcu);
408 spin_lock(&kvm->mmu_lock);
409 young = kvm_test_age_hva(kvm, address);
410 spin_unlock(&kvm->mmu_lock);
411 srcu_read_unlock(&kvm->srcu, idx);
412
413 return young;
414}
415
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100416static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
417 struct mm_struct *mm)
418{
419 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800420 int idx;
421
422 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300423 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800424 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100425}
426
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200427static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
428 .invalidate_page = kvm_mmu_notifier_invalidate_page,
429 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
430 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
431 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800432 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300433 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100434 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200435};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200436
437static int kvm_init_mmu_notifier(struct kvm *kvm)
438{
439 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
440 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
441}
442
443#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
444
445static int kvm_init_mmu_notifier(struct kvm *kvm)
446{
447 return 0;
448}
449
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200450#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
451
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800452static void kvm_init_memslots_id(struct kvm *kvm)
453{
454 int i;
455 struct kvm_memslots *slots = kvm->memslots;
456
457 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800458 slots->id_to_index[i] = slots->memslots[i].id = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800459}
460
Carsten Ottee08b9632012-01-04 10:25:20 +0100461static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800462{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100463 int r, i;
464 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800465
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100466 if (!kvm)
467 return ERR_PTR(-ENOMEM);
468
Carsten Ottee08b9632012-01-04 10:25:20 +0100469 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100470 if (r)
471 goto out_err_nodisable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200472
473 r = hardware_enable_all();
474 if (r)
475 goto out_err_nodisable;
476
Avi Kivity75858a82009-01-04 17:10:50 +0200477#ifdef CONFIG_HAVE_KVM_IRQCHIP
478 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300479 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200480#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800481
Alex Williamson1e702d92012-12-10 10:33:32 -0700482 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
483
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200484 r = -ENOMEM;
485 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
486 if (!kvm->memslots)
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100487 goto out_err_nosrcu;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800488 kvm_init_memslots_id(kvm);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200489 if (init_srcu_struct(&kvm->srcu))
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100490 goto out_err_nosrcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200491 for (i = 0; i < KVM_NR_BUSES; i++) {
492 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
493 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100494 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200495 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200496 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700497
OGAWA Hirofumi85722cd2011-05-11 09:28:28 +0900498 spin_lock_init(&kvm->mmu_lock);
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200499 kvm->mm = current->mm;
500 atomic_inc(&kvm->mm->mm_count);
Gregory Haskinsd34e6b12009-07-07 17:08:49 -0400501 kvm_eventfd_init(kvm);
Shaohua Li11ec2802007-07-23 14:51:37 +0800502 mutex_init(&kvm->lock);
Marcelo Tosatti60eead72009-06-04 15:08:23 -0300503 mutex_init(&kvm->irq_lock);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200504 mutex_init(&kvm->slots_lock);
Izik Eidusd39f13b2008-03-30 16:01:25 +0300505 atomic_set(&kvm->users_count, 1);
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700506
507 r = kvm_init_mmu_notifier(kvm);
508 if (r)
509 goto out_err;
510
Jan Kiszkae935b832011-02-08 12:55:33 +0100511 raw_spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000512 list_add(&kvm->vm_list, &vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100513 raw_spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100514
Avi Kivityf17abe92007-02-21 19:28:04 +0200515 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200516
517out_err:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100518 cleanup_srcu_struct(&kvm->srcu);
519out_err_nosrcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200520 hardware_disable_all();
521out_err_nodisable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200522 for (i = 0; i < KVM_NR_BUSES; i++)
523 kfree(kvm->buses[i]);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200524 kfree(kvm->memslots);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100525 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200526 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200527}
528
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900529/*
530 * Avoid using vmalloc for a small buffer.
531 * Should not be used when the size is statically known.
532 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900533void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900534{
535 if (size > PAGE_SIZE)
536 return vzalloc(size);
537 else
538 return kzalloc(size, GFP_KERNEL);
539}
540
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900541void kvm_kvfree(const void *addr)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900542{
543 if (is_vmalloc_addr(addr))
544 vfree(addr);
545 else
546 kfree(addr);
547}
548
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900549static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
550{
551 if (!memslot->dirty_bitmap)
552 return;
553
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900554 kvm_kvfree(memslot->dirty_bitmap);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900555 memslot->dirty_bitmap = NULL;
556}
557
Avi Kivity6aa8b732006-12-10 02:21:36 -0800558/*
559 * Free any memory in @free but not in @dont.
560 */
561static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
562 struct kvm_memory_slot *dont)
563{
Avi Kivity6aa8b732006-12-10 02:21:36 -0800564 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900565 kvm_destroy_dirty_bitmap(free);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800566
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900567 kvm_arch_free_memslot(free, dont);
Marcelo Tosatti05da4552008-02-23 11:44:30 -0300568
Avi Kivity6aa8b732006-12-10 02:21:36 -0800569 free->npages = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800570}
571
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800572void kvm_free_physmem(struct kvm *kvm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800573{
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200574 struct kvm_memslots *slots = kvm->memslots;
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800575 struct kvm_memory_slot *memslot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800576
Xiao Guangrongbe6ba0f2011-11-24 17:39:18 +0800577 kvm_for_each_memslot(memslot, slots)
578 kvm_free_physmem_slot(memslot, NULL);
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200579
580 kfree(kvm->memslots);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800581}
582
Avi Kivityf17abe92007-02-21 19:28:04 +0200583static void kvm_destroy_vm(struct kvm *kvm)
584{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200585 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200586 struct mm_struct *mm = kvm->mm;
587
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800588 kvm_arch_sync_events(kvm);
Jan Kiszkae935b832011-02-08 12:55:33 +0100589 raw_spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800590 list_del(&kvm->vm_list);
Jan Kiszkae935b832011-02-08 12:55:33 +0100591 raw_spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200592 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200593 for (i = 0; i < KVM_NR_BUSES; i++)
594 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200595 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200596#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200598#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300599 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200600#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800601 kvm_arch_destroy_vm(kvm);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100602 kvm_free_physmem(kvm);
603 cleanup_srcu_struct(&kvm->srcu);
604 kvm_arch_free_vm(kvm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200605 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200606 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200607}
608
Izik Eidusd39f13b2008-03-30 16:01:25 +0300609void kvm_get_kvm(struct kvm *kvm)
610{
611 atomic_inc(&kvm->users_count);
612}
613EXPORT_SYMBOL_GPL(kvm_get_kvm);
614
615void kvm_put_kvm(struct kvm *kvm)
616{
617 if (atomic_dec_and_test(&kvm->users_count))
618 kvm_destroy_vm(kvm);
619}
620EXPORT_SYMBOL_GPL(kvm_put_kvm);
621
622
Avi Kivityf17abe92007-02-21 19:28:04 +0200623static int kvm_vm_release(struct inode *inode, struct file *filp)
624{
625 struct kvm *kvm = filp->private_data;
626
Gregory Haskins721eecb2009-05-20 10:30:49 -0400627 kvm_irqfd_release(kvm);
628
Izik Eidusd39f13b2008-03-30 16:01:25 +0300629 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800630 return 0;
631}
632
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900633/*
634 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900635 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900636 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900637static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
638{
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900639#ifndef CONFIG_S390
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900640 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900641
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900642 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900643 if (!memslot->dirty_bitmap)
644 return -ENOMEM;
645
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900646#endif /* !CONFIG_S390 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900647 return 0;
648}
649
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800650static int cmp_memslot(const void *slot1, const void *slot2)
651{
652 struct kvm_memory_slot *s1, *s2;
653
654 s1 = (struct kvm_memory_slot *)slot1;
655 s2 = (struct kvm_memory_slot *)slot2;
656
657 if (s1->npages < s2->npages)
658 return 1;
659 if (s1->npages > s2->npages)
660 return -1;
661
662 return 0;
663}
664
665/*
666 * Sort the memslots base on its size, so the larger slots
667 * will get better fit.
668 */
669static void sort_memslots(struct kvm_memslots *slots)
670{
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800671 int i;
672
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800673 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
674 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800675
676 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
677 slots->id_to_index[slots->memslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800678}
679
Alex Williamson116c14c2012-12-21 08:20:16 -0700680void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
681 u64 last_generation)
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800682{
683 if (new) {
684 int id = new->id;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800685 struct kvm_memory_slot *old = id_to_memslot(slots, id);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800686 unsigned long npages = old->npages;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800687
Xiao Guangrong28a37542011-11-24 19:04:35 +0800688 *old = *new;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800689 if (new->npages != npages)
690 sort_memslots(slots);
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800691 }
692
Alex Williamson116c14c2012-12-21 08:20:16 -0700693 slots->generation = last_generation + 1;
Xiao Guangrongbe593d62011-11-24 17:38:24 +0800694}
695
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800696static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
697{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800698 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
699
700#ifdef KVM_CAP_READONLY_MEM
701 valid_flags |= KVM_MEM_READONLY;
702#endif
703
704 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800705 return -EINVAL;
706
707 return 0;
708}
709
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200710static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
711 struct kvm_memslots *slots, struct kvm_memory_slot *new)
712{
713 struct kvm_memslots *old_memslots = kvm->memslots;
714
715 update_memslots(slots, new, kvm->memslots->generation);
716 rcu_assign_pointer(kvm->memslots, slots);
717 synchronize_srcu_expedited(&kvm->srcu);
718 return old_memslots;
719}
720
Avi Kivity6aa8b732006-12-10 02:21:36 -0800721/*
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900722 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
723 * - create a new memory slot
724 * - delete an existing memory slot
725 * - modify an existing memory slot
726 * -- move it in the guest physical memory space
727 * -- just change its flags
728 *
729 * Since flags can be changed by some of these operations, the following
730 * differentiation is the best we can do for __kvm_set_memory_region():
731 */
732enum kvm_mr_change {
733 KVM_MR_CREATE,
734 KVM_MR_DELETE,
735 KVM_MR_MOVE,
736 KVM_MR_FLAGS_ONLY,
737};
738
739/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800740 * Allocate some memory and give it an address in the guest physical address
741 * space.
742 *
743 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800744 *
Marcelo Tosatti10589a42007-12-20 19:18:22 -0500745 * Must be called holding mmap_sem for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800746 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800747int __kvm_set_memory_region(struct kvm *kvm,
748 struct kvm_userspace_memory_region *mem,
Alex Williamsonf82a8cf2012-12-10 10:33:21 -0700749 bool user_alloc)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800750{
Avi Kivity8234b222010-12-27 12:08:45 +0200751 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800752 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200753 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900754 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800755 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700756 struct kvm_memslots *slots = NULL, *old_memslots;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900757 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800758
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800759 r = check_memory_region_flags(mem);
760 if (r)
761 goto out;
762
Avi Kivity6aa8b732006-12-10 02:21:36 -0800763 r = -EINVAL;
764 /* General sanity checks */
765 if (mem->memory_size & (PAGE_SIZE - 1))
766 goto out;
767 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
768 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900769 /* We can read the guest memory with __xxx_user() later on. */
770 if (user_alloc &&
771 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200772 !access_ok(VERIFY_WRITE,
773 (void __user *)(unsigned long)mem->userspace_addr,
774 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600775 goto out;
Xiao Guangrong93a5cef2011-11-24 17:37:48 +0800776 if (mem->slot >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800777 goto out;
778 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
779 goto out;
780
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900781 slot = id_to_memslot(kvm->memslots, mem->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800782 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
783 npages = mem->memory_size >> PAGE_SHIFT;
784
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900785 r = -EINVAL;
786 if (npages > KVM_MEM_MAX_NR_PAGES)
787 goto out;
788
Avi Kivity6aa8b732006-12-10 02:21:36 -0800789 if (!npages)
790 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
791
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900792 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800793
Avi Kivitye36d96f2010-06-21 10:56:36 +0300794 new.id = mem->slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800795 new.base_gfn = base_gfn;
796 new.npages = npages;
797 new.flags = mem->flags;
798
Avi Kivity6aa8b732006-12-10 02:21:36 -0800799 r = -EINVAL;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900800 if (npages) {
801 if (!old.npages)
802 change = KVM_MR_CREATE;
803 else { /* Modify an existing slot. */
804 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900805 (npages != old.npages) ||
806 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900807 goto out;
808
809 if (base_gfn != old.base_gfn)
810 change = KVM_MR_MOVE;
811 else if (new.flags != old.flags)
812 change = KVM_MR_FLAGS_ONLY;
813 else { /* Nothing to change. */
814 r = 0;
815 goto out;
816 }
817 }
818 } else if (old.npages) {
819 change = KVM_MR_DELETE;
820 } else /* Modify a non-existent slot: disallowed. */
Takuya Yoshikawa0ea75e12013-01-11 18:26:10 +0900821 goto out;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800822
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900823 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900824 /* Check for overlaps */
825 r = -EEXIST;
826 kvm_for_each_memslot(slot, kvm->memslots) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900827 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
828 (slot->id == mem->slot))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900829 continue;
830 if (!((base_gfn + npages <= slot->base_gfn) ||
831 (base_gfn >= slot->base_gfn + slot->npages)))
832 goto out;
833 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800834 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800835
Avi Kivity6aa8b732006-12-10 02:21:36 -0800836 /* Free page dirty bitmap if unneeded */
837 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000838 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800839
840 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900841 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900842 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900843
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900844 if (kvm_arch_create_memslot(&new, npages))
845 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800846 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200847
Avi Kivity6aa8b732006-12-10 02:21:36 -0800848 /* Allocate page dirty bitmap if needed */
849 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900850 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800851 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800852 }
853
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900854 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200855 r = -ENOMEM;
Thomas Meyer6da64fd2011-11-08 20:32:19 +0100856 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
857 GFP_KERNEL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200858 if (!slots)
859 goto out_free;
Xiao Guangrong28a37542011-11-24 19:04:35 +0800860 slot = id_to_memslot(slots, mem->slot);
861 slot->flags |= KVM_MEMSLOT_INVALID;
862
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200863 old_memslots = install_new_memslots(kvm, slots, NULL);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200864
Alex Williamsone40f1932012-12-10 10:32:57 -0700865 /* slot was deleted or moved, clear iommu mapping */
866 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300867 /* From this point no new shadow pages pointing to a deleted,
868 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200869 *
870 * validation of sp->gfn happens in:
871 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
872 * - kvm_is_visible_gfn (mmu_check_roots)
873 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300874 kvm_arch_flush_shadow_memslot(kvm, slot);
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700875 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200876 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -0300877
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200878 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
879 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700880 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200881
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200882 r = -ENOMEM;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700883 /*
884 * We can re-use the old_memslots from above, the only difference
885 * from the currently installed memslots is the invalid flag. This
886 * will get overwritten by update_memslots anyway.
887 */
888 if (!slots) {
889 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
890 GFP_KERNEL);
891 if (!slots)
892 goto out_free;
893 }
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200894
Alex Williamson261874b2013-01-24 15:04:03 -0700895 /*
896 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900897 * un-mapped and re-mapped if their base changes. Since base change
898 * unmapping is handled above with slot deletion, mapping alone is
899 * needed here. Anything else the iommu might care about for existing
900 * slots (size changes, userspace addr changes and read-only flag
901 * changes) is disallowed above, so any other attribute changes getting
902 * here can be skipped.
Alex Williamson261874b2013-01-24 15:04:03 -0700903 */
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900904 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
905 r = kvm_iommu_map_pages(kvm, &new);
906 if (r)
907 goto out_slots;
Alex Williamsone40f1932012-12-10 10:32:57 -0700908 }
909
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200910 /* actual memory is freed via old in kvm_free_physmem_slot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900911 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200912 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900913 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200914 }
915
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200916 old_memslots = install_new_memslots(kvm, slots, &new);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800917
Marcelo Tosattif7784b82009-12-23 14:35:18 -0200918 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +0800919
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200920 kvm_free_physmem_slot(&old, &new);
921 kfree(old_memslots);
922
Avi Kivity6aa8b732006-12-10 02:21:36 -0800923 return 0;
924
Alex Williamsone40f1932012-12-10 10:32:57 -0700925out_slots:
926 kfree(slots);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800927out_free:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800928 kvm_free_physmem_slot(&new, &old);
929out:
930 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +0200931}
Sheng Yangf78e0e22007-10-29 09:40:42 +0800932EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
933
934int kvm_set_memory_region(struct kvm *kvm,
935 struct kvm_userspace_memory_region *mem,
Alex Williamsonf82a8cf2012-12-10 10:33:21 -0700936 bool user_alloc)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800937{
938 int r;
939
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200940 mutex_lock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800941 r = __kvm_set_memory_region(kvm, mem, user_alloc);
Marcelo Tosatti79fac952009-12-23 14:35:26 -0200942 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +0800943 return r;
944}
Izik Eidus210c7c42007-10-24 23:52:57 +0200945EXPORT_SYMBOL_GPL(kvm_set_memory_region);
946
Carsten Otte1fe779f2007-10-29 16:08:35 +0100947int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
948 struct
949 kvm_userspace_memory_region *mem,
Alex Williamsonf82a8cf2012-12-10 10:33:21 -0700950 bool user_alloc)
Izik Eidus210c7c42007-10-24 23:52:57 +0200951{
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700952 if (mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +0200953 return -EINVAL;
Izik Eidus210c7c42007-10-24 23:52:57 +0200954 return kvm_set_memory_region(kvm, mem, user_alloc);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800955}
956
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800957int kvm_get_dirty_log(struct kvm *kvm,
958 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800959{
960 struct kvm_memory_slot *memslot;
961 int r, i;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900962 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800963 unsigned long any = 0;
964
Avi Kivity6aa8b732006-12-10 02:21:36 -0800965 r = -EINVAL;
Alex Williamsonbbacc0c2012-12-10 10:33:09 -0700966 if (log->slot >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967 goto out;
968
Xiao Guangrong28a37542011-11-24 19:04:35 +0800969 memslot = id_to_memslot(kvm->memslots, log->slot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800970 r = -ENOENT;
971 if (!memslot->dirty_bitmap)
972 goto out;
973
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +0900974 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800975
Uri Lublincd1a4a92007-02-22 16:43:09 +0200976 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977 any = memslot->dirty_bitmap[i];
978
979 r = -EFAULT;
980 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
981 goto out;
982
Zhang Xiantao5bb064d2007-11-18 20:29:43 +0800983 if (any)
984 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800985
986 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800987out:
Avi Kivity6aa8b732006-12-10 02:21:36 -0800988 return r;
989}
990
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900991bool kvm_largepages_enabled(void)
992{
993 return largepages_enabled;
994}
995
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300996void kvm_disable_largepages(void)
997{
998 largepages_enabled = false;
999}
1000EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1001
Gleb Natapov49c77542010-10-18 15:22:23 +02001002struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1003{
1004 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1005}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001006EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001007
Izik Eiduse0d62c72007-10-24 23:57:46 +02001008int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1009{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001010 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001011
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001012 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001013 memslot->flags & KVM_MEMSLOT_INVALID)
1014 return 0;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001015
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001016 return 1;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001017}
1018EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1019
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001020unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1021{
1022 struct vm_area_struct *vma;
1023 unsigned long addr, size;
1024
1025 size = PAGE_SIZE;
1026
1027 addr = gfn_to_hva(kvm, gfn);
1028 if (kvm_is_error_hva(addr))
1029 return PAGE_SIZE;
1030
1031 down_read(&current->mm->mmap_sem);
1032 vma = find_vma(current->mm, addr);
1033 if (!vma)
1034 goto out;
1035
1036 size = vma_kernel_pagesize(vma);
1037
1038out:
1039 up_read(&current->mm->mmap_sem);
1040
1041 return size;
1042}
1043
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001044static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1045{
1046 return slot->flags & KVM_MEM_READONLY;
1047}
1048
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001049static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1050 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001051{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001052 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001053 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001054
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001055 if (memslot_is_readonly(slot) && write)
1056 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001057
1058 if (nr_pages)
1059 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1060
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001061 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001062}
Xiao Guangrong48987782010-08-22 19:11:43 +08001063
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001064static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1065 gfn_t *nr_pages)
1066{
1067 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1068}
1069
1070unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1071 gfn_t gfn)
1072{
1073 return gfn_to_hva_many(slot, gfn, NULL);
1074}
1075EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1076
Xiao Guangrong48987782010-08-22 19:11:43 +08001077unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1078{
Gleb Natapov49c77542010-10-18 15:22:23 +02001079 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001080}
Sheng Yang0d150292008-04-25 21:44:50 +08001081EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001082
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001083/*
1084 * The hva returned by this function is only allowed to be read.
1085 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1086 */
1087static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
Gleb Natapov80300892010-10-19 18:13:41 +02001088{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001089 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001090}
1091
1092static int kvm_read_hva(void *data, void __user *hva, int len)
1093{
1094 return __copy_from_user(data, hva, len);
1095}
1096
1097static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1098{
1099 return __copy_from_user_inatomic(data, hva, len);
Gleb Natapov80300892010-10-19 18:13:41 +02001100}
1101
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001102int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1103 unsigned long start, int write, struct page **page)
1104{
1105 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1106
1107 if (write)
1108 flags |= FOLL_WRITE;
1109
1110 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1111}
1112
Huang Yingfafc3db2011-01-30 11:15:49 +08001113static inline int check_user_page_hwpoison(unsigned long addr)
1114{
1115 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1116
1117 rc = __get_user_pages(current, current->mm, addr, 1,
1118 flags, NULL, NULL, NULL);
1119 return rc == -EHWPOISON;
1120}
1121
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001122/*
1123 * The atomic path to get the writable pfn which will be stored in @pfn,
1124 * true indicates success, otherwise false is returned.
1125 */
1126static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1127 bool write_fault, bool *writable, pfn_t *pfn)
1128{
1129 struct page *page[1];
1130 int npages;
1131
1132 if (!(async || atomic))
1133 return false;
1134
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001135 /*
1136 * Fast pin a writable pfn only if it is a write fault request
1137 * or the caller allows to map a writable pfn for a read fault
1138 * request.
1139 */
1140 if (!(write_fault || writable))
1141 return false;
1142
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001143 npages = __get_user_pages_fast(addr, 1, 1, page);
1144 if (npages == 1) {
1145 *pfn = page_to_pfn(page[0]);
1146
1147 if (writable)
1148 *writable = true;
1149 return true;
1150 }
1151
1152 return false;
1153}
1154
1155/*
1156 * The slow path to get the pfn of the specified host virtual address,
1157 * 1 indicates success, -errno is returned if error is detected.
1158 */
1159static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1160 bool *writable, pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001161{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001162 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001163 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001164
1165 might_sleep();
1166
1167 if (writable)
1168 *writable = write_fault;
1169
1170 if (async) {
1171 down_read(&current->mm->mmap_sem);
1172 npages = get_user_page_nowait(current, current->mm,
1173 addr, write_fault, page);
1174 up_read(&current->mm->mmap_sem);
1175 } else
1176 npages = get_user_pages_fast(addr, 1, write_fault,
1177 page);
1178 if (npages != 1)
1179 return npages;
1180
1181 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001182 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001183 struct page *wpage[1];
1184
1185 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1186 if (npages == 1) {
1187 *writable = true;
1188 put_page(page[0]);
1189 page[0] = wpage[0];
1190 }
1191
1192 npages = 1;
1193 }
1194 *pfn = page_to_pfn(page[0]);
1195 return npages;
1196}
1197
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001198static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1199{
1200 if (unlikely(!(vma->vm_flags & VM_READ)))
1201 return false;
1202
1203 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1204 return false;
1205
1206 return true;
1207}
1208
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001209/*
1210 * Pin guest page in memory and return its pfn.
1211 * @addr: host virtual address which maps memory to the guest
1212 * @atomic: whether this function can sleep
1213 * @async: whether this function need to wait IO complete if the
1214 * host page is not in the memory
1215 * @write_fault: whether we should get a writable host page
1216 * @writable: whether it allows to map a writable host page for !@write_fault
1217 *
1218 * The function will map a writable host page for these two cases:
1219 * 1): @write_fault = true
1220 * 2): @write_fault = false && @writable, @writable will tell the caller
1221 * whether the mapping is writable.
1222 */
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001223static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1224 bool write_fault, bool *writable)
1225{
1226 struct vm_area_struct *vma;
1227 pfn_t pfn = 0;
1228 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001229
Gleb Natapovaf585b92010-10-14 11:22:46 +02001230 /* we can do it either atomically or asynchronously, not both */
1231 BUG_ON(atomic && async);
1232
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001233 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1234 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001235
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001236 if (atomic)
1237 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001238
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001239 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1240 if (npages == 1)
1241 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001242
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001243 down_read(&current->mm->mmap_sem);
1244 if (npages == -EHWPOISON ||
1245 (!async && check_user_page_hwpoison(addr))) {
1246 pfn = KVM_PFN_ERR_HWPOISON;
1247 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001248 }
Izik Eidus539cb662007-11-11 22:05:04 +02001249
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001250 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001251
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001252 if (vma == NULL)
1253 pfn = KVM_PFN_ERR_FAULT;
1254 else if ((vma->vm_flags & VM_PFNMAP)) {
1255 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1256 vma->vm_pgoff;
1257 BUG_ON(!kvm_is_mmio_pfn(pfn));
1258 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001259 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001260 *async = true;
1261 pfn = KVM_PFN_ERR_FAULT;
1262 }
1263exit:
1264 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001265 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001266}
1267
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001268static pfn_t
1269__gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1270 bool *async, bool write_fault, bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001271{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001272 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1273
1274 if (addr == KVM_HVA_ERR_RO_BAD)
1275 return KVM_PFN_ERR_RO_FAULT;
1276
1277 if (kvm_is_error_hva(addr))
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001278 return KVM_PFN_NOSLOT;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001279
1280 /* Do not map writable pfn in the readonly memslot. */
1281 if (writable && memslot_is_readonly(slot)) {
1282 *writable = false;
1283 writable = NULL;
1284 }
1285
1286 return hva_to_pfn(addr, atomic, async, write_fault,
1287 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001288}
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001289
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001290static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1291 bool write_fault, bool *writable)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001292{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001293 struct kvm_memory_slot *slot;
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001294
Gleb Natapovaf585b92010-10-14 11:22:46 +02001295 if (async)
1296 *async = false;
1297
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001298 slot = gfn_to_memslot(kvm, gfn);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001299
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001300 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1301 writable);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001302}
1303
1304pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1305{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001306 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001307}
1308EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1309
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001310pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1311 bool write_fault, bool *writable)
Gleb Natapovaf585b92010-10-14 11:22:46 +02001312{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001313 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
Gleb Natapovaf585b92010-10-14 11:22:46 +02001314}
1315EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1316
Xiao Guangrong365fb3f2010-08-28 19:24:13 +08001317pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1318{
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001319 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001320}
Anthony Liguori35149e22008-04-02 14:46:56 -05001321EXPORT_SYMBOL_GPL(gfn_to_pfn);
1322
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001323pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1324 bool *writable)
1325{
1326 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1327}
1328EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1329
Xiao Guangrongd5661042012-07-17 21:56:16 +08001330pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001331{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001332 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001333}
1334
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001335pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1336{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001337 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001338}
1339EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1340
Xiao Guangrong48987782010-08-22 19:11:43 +08001341int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1342 int nr_pages)
1343{
1344 unsigned long addr;
1345 gfn_t entry;
1346
Gleb Natapov49c77542010-10-18 15:22:23 +02001347 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001348 if (kvm_is_error_hva(addr))
1349 return -1;
1350
1351 if (entry < nr_pages)
1352 return 0;
1353
1354 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1355}
1356EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1357
Xiao Guangronga2766322012-07-26 11:58:59 +08001358static struct page *kvm_pfn_to_page(pfn_t pfn)
1359{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001360 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001361 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001362
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001363 if (kvm_is_mmio_pfn(pfn)) {
1364 WARN_ON(1);
1365 return KVM_ERR_PTR_BAD_PAGE;
1366 }
1367
Xiao Guangronga2766322012-07-26 11:58:59 +08001368 return pfn_to_page(pfn);
1369}
1370
Anthony Liguori35149e22008-04-02 14:46:56 -05001371struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1372{
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001373 pfn_t pfn;
1374
1375 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001376
Xiao Guangronga2766322012-07-26 11:58:59 +08001377 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001378}
Anthony Liguoriaab61cc2007-10-29 15:15:20 -05001379
Avi Kivity954bbbc2007-03-30 14:02:32 +03001380EXPORT_SYMBOL_GPL(gfn_to_page);
1381
Izik Eidusb4231d62007-11-20 11:49:33 +02001382void kvm_release_page_clean(struct page *page)
1383{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001384 WARN_ON(is_error_page(page));
1385
Anthony Liguori35149e22008-04-02 14:46:56 -05001386 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001387}
1388EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1389
Anthony Liguori35149e22008-04-02 14:46:56 -05001390void kvm_release_pfn_clean(pfn_t pfn)
1391{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001392 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001393 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001394}
1395EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1396
Izik Eidusb4231d62007-11-20 11:49:33 +02001397void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001398{
Xiao Guangronga2766322012-07-26 11:58:59 +08001399 WARN_ON(is_error_page(page));
1400
Anthony Liguori35149e22008-04-02 14:46:56 -05001401 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001402}
Izik Eidusb4231d62007-11-20 11:49:33 +02001403EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001404
Anthony Liguori35149e22008-04-02 14:46:56 -05001405void kvm_release_pfn_dirty(pfn_t pfn)
1406{
1407 kvm_set_pfn_dirty(pfn);
1408 kvm_release_pfn_clean(pfn);
1409}
1410EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1411
1412void kvm_set_page_dirty(struct page *page)
1413{
1414 kvm_set_pfn_dirty(page_to_pfn(page));
1415}
1416EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1417
1418void kvm_set_pfn_dirty(pfn_t pfn)
1419{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001420 if (!kvm_is_mmio_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001421 struct page *page = pfn_to_page(pfn);
1422 if (!PageReserved(page))
1423 SetPageDirty(page);
1424 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001425}
1426EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1427
1428void kvm_set_pfn_accessed(pfn_t pfn)
1429{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001430 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001431 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001432}
1433EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1434
1435void kvm_get_pfn(pfn_t pfn)
1436{
Xiantao Zhangc77fb9d2008-09-27 10:55:40 +08001437 if (!kvm_is_mmio_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001438 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001439}
1440EXPORT_SYMBOL_GPL(kvm_get_pfn);
1441
Izik Eidus195aefd2007-10-01 22:14:18 +02001442static int next_segment(unsigned long len, int offset)
1443{
1444 if (len > PAGE_SIZE - offset)
1445 return PAGE_SIZE - offset;
1446 else
1447 return len;
1448}
1449
1450int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1451 int len)
1452{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001453 int r;
1454 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001455
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001456 addr = gfn_to_hva_read(kvm, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001457 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001458 return -EFAULT;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001459 r = kvm_read_hva(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001460 if (r)
1461 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001462 return 0;
1463}
1464EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1465
1466int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1467{
1468 gfn_t gfn = gpa >> PAGE_SHIFT;
1469 int seg;
1470 int offset = offset_in_page(gpa);
1471 int ret;
1472
1473 while ((seg = next_segment(len, offset)) != 0) {
1474 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1475 if (ret < 0)
1476 return ret;
1477 offset = 0;
1478 len -= seg;
1479 data += seg;
1480 ++gfn;
1481 }
1482 return 0;
1483}
1484EXPORT_SYMBOL_GPL(kvm_read_guest);
1485
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001486int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1487 unsigned long len)
1488{
1489 int r;
1490 unsigned long addr;
1491 gfn_t gfn = gpa >> PAGE_SHIFT;
1492 int offset = offset_in_page(gpa);
1493
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001494 addr = gfn_to_hva_read(kvm, gfn);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001495 if (kvm_is_error_hva(addr))
1496 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001497 pagefault_disable();
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001498 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001499 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001500 if (r)
1501 return -EFAULT;
1502 return 0;
1503}
1504EXPORT_SYMBOL(kvm_read_guest_atomic);
1505
Izik Eidus195aefd2007-10-01 22:14:18 +02001506int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1507 int offset, int len)
1508{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001509 int r;
1510 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001511
Izik Eiduse0506bc2007-11-11 22:10:22 +02001512 addr = gfn_to_hva(kvm, gfn);
1513 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001514 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001515 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001516 if (r)
1517 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001518 mark_page_dirty(kvm, gfn);
1519 return 0;
1520}
1521EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1522
1523int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1524 unsigned long len)
1525{
1526 gfn_t gfn = gpa >> PAGE_SHIFT;
1527 int seg;
1528 int offset = offset_in_page(gpa);
1529 int ret;
1530
1531 while ((seg = next_segment(len, offset)) != 0) {
1532 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1533 if (ret < 0)
1534 return ret;
1535 offset = 0;
1536 len -= seg;
1537 data += seg;
1538 ++gfn;
1539 }
1540 return 0;
1541}
1542
Gleb Natapov49c77542010-10-18 15:22:23 +02001543int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1544 gpa_t gpa)
1545{
1546 struct kvm_memslots *slots = kvm_memslots(kvm);
1547 int offset = offset_in_page(gpa);
1548 gfn_t gfn = gpa >> PAGE_SHIFT;
1549
1550 ghc->gpa = gpa;
1551 ghc->generation = slots->generation;
Paul Mackerras9d4cba72012-01-12 20:09:51 +00001552 ghc->memslot = gfn_to_memslot(kvm, gfn);
Gleb Natapov49c77542010-10-18 15:22:23 +02001553 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1554 if (!kvm_is_error_hva(ghc->hva))
1555 ghc->hva += offset;
1556 else
1557 return -EFAULT;
1558
1559 return 0;
1560}
1561EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1562
1563int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1564 void *data, unsigned long len)
1565{
1566 struct kvm_memslots *slots = kvm_memslots(kvm);
1567 int r;
1568
1569 if (slots->generation != ghc->generation)
1570 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1571
1572 if (kvm_is_error_hva(ghc->hva))
1573 return -EFAULT;
1574
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001575 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001576 if (r)
1577 return -EFAULT;
1578 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1579
1580 return 0;
1581}
1582EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1583
Gleb Natapove03b6442011-07-11 15:28:11 -04001584int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1585 void *data, unsigned long len)
1586{
1587 struct kvm_memslots *slots = kvm_memslots(kvm);
1588 int r;
1589
1590 if (slots->generation != ghc->generation)
1591 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1592
1593 if (kvm_is_error_hva(ghc->hva))
1594 return -EFAULT;
1595
1596 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1597 if (r)
1598 return -EFAULT;
1599
1600 return 0;
1601}
1602EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1603
Izik Eidus195aefd2007-10-01 22:14:18 +02001604int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1605{
Heiko Carstens3bcc8a82010-10-27 17:21:21 +02001606 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1607 offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001608}
1609EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1610
1611int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1612{
1613 gfn_t gfn = gpa >> PAGE_SHIFT;
1614 int seg;
1615 int offset = offset_in_page(gpa);
1616 int ret;
1617
1618 while ((seg = next_segment(len, offset)) != 0) {
1619 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1620 if (ret < 0)
1621 return ret;
1622 offset = 0;
1623 len -= seg;
1624 ++gfn;
1625 }
1626 return 0;
1627}
1628EXPORT_SYMBOL_GPL(kvm_clear_guest);
1629
Gleb Natapov49c77542010-10-18 15:22:23 +02001630void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1631 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001632{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001633 if (memslot && memslot->dirty_bitmap) {
1634 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001635
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001636 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001637 }
1638}
1639
Gleb Natapov49c77542010-10-18 15:22:23 +02001640void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1641{
1642 struct kvm_memory_slot *memslot;
1643
1644 memslot = gfn_to_memslot(kvm, gfn);
1645 mark_page_dirty_in_slot(kvm, memslot, gfn);
1646}
1647
Eddie Dongb6958ce2007-07-18 12:15:21 +03001648/*
1649 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1650 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05001651void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03001652{
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001653 DEFINE_WAIT(wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001654
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001655 for (;;) {
1656 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001657
Gleb Natapova1b37102009-07-09 15:33:52 +03001658 if (kvm_arch_vcpu_runnable(vcpu)) {
Avi Kivitya8eeb042010-05-10 12:34:53 +03001659 kvm_make_request(KVM_REQ_UNHALT, vcpu);
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001660 break;
Marcelo Tosattid7690172008-09-08 15:23:48 -03001661 }
Gleb Natapov09cec752009-03-23 15:11:44 +02001662 if (kvm_cpu_has_pending_timer(vcpu))
1663 break;
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001664 if (signal_pending(current))
1665 break;
1666
Eddie Dongb6958ce2007-07-18 12:15:21 +03001667 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03001668 }
1669
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03001670 finish_wait(&vcpu->wq, &wait);
Eddie Dongb6958ce2007-07-18 12:15:21 +03001671}
1672
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001673#ifndef CONFIG_S390
Christoffer Dallb6d33832012-03-08 16:44:24 -05001674/*
1675 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1676 */
1677void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1678{
1679 int me;
1680 int cpu = vcpu->cpu;
1681 wait_queue_head_t *wqp;
1682
1683 wqp = kvm_arch_vcpu_wq(vcpu);
1684 if (waitqueue_active(wqp)) {
1685 wake_up_interruptible(wqp);
1686 ++vcpu->stat.halt_wakeup;
1687 }
1688
1689 me = get_cpu();
1690 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1691 if (kvm_arch_vcpu_should_kick(vcpu))
1692 smp_send_reschedule(cpu);
1693 put_cpu();
1694}
Marcelo Tosatti8c847802012-03-14 17:58:48 -03001695#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05001696
Avi Kivity6aa8b732006-12-10 02:21:36 -08001697void kvm_resched(struct kvm_vcpu *vcpu)
1698{
Yaozu Dong3fca0362007-04-25 16:49:19 +03001699 if (!need_resched())
1700 return;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001701 cond_resched();
Avi Kivity6aa8b732006-12-10 02:21:36 -08001702}
1703EXPORT_SYMBOL_GPL(kvm_resched);
1704
Konstantin Weitz41628d32012-04-25 15:30:38 +02001705bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1706{
1707 struct pid *pid;
1708 struct task_struct *task = NULL;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301709 bool ret = false;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001710
1711 rcu_read_lock();
1712 pid = rcu_dereference(target->pid);
1713 if (pid)
1714 task = get_pid_task(target->pid, PIDTYPE_PID);
1715 rcu_read_unlock();
1716 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301717 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001718 if (task->flags & PF_VCPU) {
1719 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301720 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001721 }
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301722 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02001723 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301724
1725 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02001726}
1727EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1728
Raghavendra K T06e48c52012-07-19 15:17:52 +05301729#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1730/*
1731 * Helper that checks whether a VCPU is eligible for directed yield.
1732 * Most eligible candidate to yield is decided by following heuristics:
1733 *
1734 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1735 * (preempted lock holder), indicated by @in_spin_loop.
1736 * Set at the beiginning and cleared at the end of interception/PLE handler.
1737 *
1738 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1739 * chance last time (mostly it has become eligible now since we have probably
1740 * yielded to lockholder in last iteration. This is done by toggling
1741 * @dy_eligible each time a VCPU checked for eligibility.)
1742 *
1743 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1744 * to preempted lock-holder could result in wrong VCPU selection and CPU
1745 * burning. Giving priority for a potential lock-holder increases lock
1746 * progress.
1747 *
1748 * Since algorithm is based on heuristics, accessing another VCPU data without
1749 * locking does not harm. It may result in trying to yield to same VCPU, fail
1750 * and continue with next VCPU and so on.
1751 */
1752bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1753{
1754 bool eligible;
1755
1756 eligible = !vcpu->spin_loop.in_spin_loop ||
1757 (vcpu->spin_loop.in_spin_loop &&
1758 vcpu->spin_loop.dy_eligible);
1759
1760 if (vcpu->spin_loop.in_spin_loop)
1761 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1762
1763 return eligible;
1764}
1765#endif
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301766
Rik van Riel217ece62011-02-01 09:53:28 -05001767void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001768{
Rik van Riel217ece62011-02-01 09:53:28 -05001769 struct kvm *kvm = me->kvm;
1770 struct kvm_vcpu *vcpu;
1771 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1772 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301773 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05001774 int pass;
1775 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001776
Raghavendra K T4c088492012-07-18 19:07:46 +05301777 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05001778 /*
1779 * We boost the priority of a VCPU that is runnable but not
1780 * currently running, because it got preempted by something
1781 * else and called schedule in __vcpu_run. Hopefully that
1782 * VCPU is holding the lock that we need and will release it.
1783 * We approximate round-robin by starting at the last boosted VCPU.
1784 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301785 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05001786 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04001787 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05001788 i = last_boosted_vcpu;
1789 continue;
1790 } else if (pass && i > last_boosted_vcpu)
1791 break;
1792 if (vcpu == me)
1793 continue;
1794 if (waitqueue_active(&vcpu->wq))
1795 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05301796 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1797 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301798
1799 yielded = kvm_vcpu_yield_to(vcpu);
1800 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05001801 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05001802 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05301803 } else if (yielded < 0) {
1804 try--;
1805 if (!try)
1806 break;
Rik van Riel217ece62011-02-01 09:53:28 -05001807 }
Rik van Riel217ece62011-02-01 09:53:28 -05001808 }
1809 }
Raghavendra K T4c088492012-07-18 19:07:46 +05301810 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05301811
1812 /* Ensure vcpu is not eligible during next spinloop */
1813 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08001814}
1815EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1816
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001817static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001818{
1819 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001820 struct page *page;
1821
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001822 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02001823 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02001824#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001825 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001826 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02001827#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02001828#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1829 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1830 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1831#endif
Avi Kivity039576c2007-03-20 12:46:50 +02001832 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01001833 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001834 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001835 vmf->page = page;
1836 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001837}
1838
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001839static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11001840 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001841};
1842
1843static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1844{
1845 vma->vm_ops = &kvm_vcpu_vm_ops;
1846 return 0;
1847}
1848
Avi Kivitybccf2152007-02-21 18:04:26 +02001849static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1850{
1851 struct kvm_vcpu *vcpu = filp->private_data;
1852
Al Viro66c0b392008-04-19 20:33:56 +01001853 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001854 return 0;
1855}
1856
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01001857static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02001858 .release = kvm_vcpu_release,
1859 .unlocked_ioctl = kvm_vcpu_ioctl,
Alexander Graf1dda6062011-06-08 02:45:37 +02001860#ifdef CONFIG_COMPAT
1861 .compat_ioctl = kvm_vcpu_compat_ioctl,
1862#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001863 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001864 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02001865};
1866
1867/*
1868 * Allocates an inode for the vcpu.
1869 */
1870static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1871{
Roland Dreier628ff7c2009-12-18 09:41:24 -08001872 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
Avi Kivitybccf2152007-02-21 18:04:26 +02001873}
1874
Avi Kivityc5ea7662007-02-20 18:41:05 +02001875/*
1876 * Creates some virtual cpus. Good luck creating more than one.
1877 */
Gleb Natapov73880c82009-06-09 15:56:28 +03001878static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02001879{
1880 int r;
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001881 struct kvm_vcpu *vcpu, *v;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001882
Gleb Natapov73880c82009-06-09 15:56:28 +03001883 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001884 if (IS_ERR(vcpu))
1885 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001886
Avi Kivity15ad7142007-07-11 18:17:21 +03001887 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1888
Avi Kivity26e52152007-11-20 15:30:24 +02001889 r = kvm_arch_vcpu_setup(vcpu);
1890 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02001891 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02001892
Shaohua Li11ec2802007-07-23 14:51:37 +08001893 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02001894 if (!kvm_vcpu_compatible(vcpu)) {
1895 r = -EINVAL;
1896 goto unlock_vcpu_destroy;
1897 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001898 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1899 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02001900 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001901 }
Gleb Natapov73880c82009-06-09 15:56:28 +03001902
Gleb Natapov988a2ca2009-06-09 15:56:29 +03001903 kvm_for_each_vcpu(r, v, kvm)
1904 if (v->vcpu_id == id) {
Gleb Natapov73880c82009-06-09 15:56:28 +03001905 r = -EEXIST;
Jan Kiszkad7805922011-05-23 10:33:05 +02001906 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001907 }
1908
1909 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001910
1911 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01001912 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02001913 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03001914 if (r < 0) {
1915 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02001916 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03001917 }
1918
1919 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1920 smp_wmb();
1921 atomic_inc(&kvm->online_vcpus);
1922
Gleb Natapov73880c82009-06-09 15:56:28 +03001923 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02001924 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02001925 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02001926
Jan Kiszkad7805922011-05-23 10:33:05 +02001927unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03001928 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02001929vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06001930 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02001931 return r;
1932}
1933
Avi Kivity1961d272007-03-05 19:46:05 +02001934static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1935{
1936 if (sigset) {
1937 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1938 vcpu->sigset_active = 1;
1939 vcpu->sigset = *sigset;
1940 } else
1941 vcpu->sigset_active = 0;
1942 return 0;
1943}
1944
Avi Kivitybccf2152007-02-21 18:04:26 +02001945static long kvm_vcpu_ioctl(struct file *filp,
1946 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001947{
Avi Kivitybccf2152007-02-21 18:04:26 +02001948 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00001949 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02001950 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07001951 struct kvm_fpu *fpu = NULL;
1952 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001953
Avi Kivity6d4e4c42007-11-21 16:41:05 +02001954 if (vcpu->kvm->mm != current->mm)
1955 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03001956
1957#if defined(CONFIG_S390) || defined(CONFIG_PPC)
1958 /*
1959 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1960 * so vcpu_load() would break it.
1961 */
1962 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1963 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1964#endif
1965
1966
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03001967 r = vcpu_load(vcpu);
1968 if (r)
1969 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001970 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02001971 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02001972 r = -EINVAL;
1973 if (arg)
1974 goto out;
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05001975 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02001976 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001977 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001978 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001979 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001980
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001981 r = -ENOMEM;
1982 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1983 if (!kvm_regs)
1984 goto out;
1985 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001986 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001987 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001988 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001989 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1990 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001991 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001992out_free1:
1993 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994 break;
1995 }
1996 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001997 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001998
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08001999 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002000 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2001 if (IS_ERR(kvm_regs)) {
2002 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002003 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002004 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002005 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002006 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002007 break;
2008 }
2009 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002010 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2011 r = -ENOMEM;
2012 if (!kvm_sregs)
2013 goto out;
2014 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002015 if (r)
2016 goto out;
2017 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002018 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002019 goto out;
2020 r = 0;
2021 break;
2022 }
2023 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002024 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2025 if (IS_ERR(kvm_sregs)) {
2026 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002027 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002028 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002029 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002030 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002031 break;
2032 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002033 case KVM_GET_MP_STATE: {
2034 struct kvm_mp_state mp_state;
2035
2036 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2037 if (r)
2038 goto out;
2039 r = -EFAULT;
2040 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2041 goto out;
2042 r = 0;
2043 break;
2044 }
2045 case KVM_SET_MP_STATE: {
2046 struct kvm_mp_state mp_state;
2047
2048 r = -EFAULT;
2049 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2050 goto out;
2051 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002052 break;
2053 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002054 case KVM_TRANSLATE: {
2055 struct kvm_translation tr;
2056
2057 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002058 if (copy_from_user(&tr, argp, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002059 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002060 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002061 if (r)
2062 goto out;
2063 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002064 if (copy_to_user(argp, &tr, sizeof tr))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002065 goto out;
2066 r = 0;
2067 break;
2068 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002069 case KVM_SET_GUEST_DEBUG: {
2070 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002071
2072 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002073 if (copy_from_user(&dbg, argp, sizeof dbg))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002074 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002075 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002076 break;
2077 }
Avi Kivity1961d272007-03-05 19:46:05 +02002078 case KVM_SET_SIGNAL_MASK: {
2079 struct kvm_signal_mask __user *sigmask_arg = argp;
2080 struct kvm_signal_mask kvm_sigmask;
2081 sigset_t sigset, *p;
2082
2083 p = NULL;
2084 if (argp) {
2085 r = -EFAULT;
2086 if (copy_from_user(&kvm_sigmask, argp,
2087 sizeof kvm_sigmask))
2088 goto out;
2089 r = -EINVAL;
2090 if (kvm_sigmask.len != sizeof sigset)
2091 goto out;
2092 r = -EFAULT;
2093 if (copy_from_user(&sigset, sigmask_arg->sigset,
2094 sizeof sigset))
2095 goto out;
2096 p = &sigset;
2097 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002098 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002099 break;
2100 }
Avi Kivityb8836732007-04-01 16:34:31 +03002101 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002102 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2103 r = -ENOMEM;
2104 if (!fpu)
2105 goto out;
2106 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002107 if (r)
2108 goto out;
2109 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002110 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002111 goto out;
2112 r = 0;
2113 break;
2114 }
2115 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002116 fpu = memdup_user(argp, sizeof(*fpu));
2117 if (IS_ERR(fpu)) {
2118 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002119 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002120 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002121 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002122 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002123 break;
2124 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002125 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002126 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002127 }
2128out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002129 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002130 kfree(fpu);
2131 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002132 return r;
2133}
2134
Alexander Graf1dda6062011-06-08 02:45:37 +02002135#ifdef CONFIG_COMPAT
2136static long kvm_vcpu_compat_ioctl(struct file *filp,
2137 unsigned int ioctl, unsigned long arg)
2138{
2139 struct kvm_vcpu *vcpu = filp->private_data;
2140 void __user *argp = compat_ptr(arg);
2141 int r;
2142
2143 if (vcpu->kvm->mm != current->mm)
2144 return -EIO;
2145
2146 switch (ioctl) {
2147 case KVM_SET_SIGNAL_MASK: {
2148 struct kvm_signal_mask __user *sigmask_arg = argp;
2149 struct kvm_signal_mask kvm_sigmask;
2150 compat_sigset_t csigset;
2151 sigset_t sigset;
2152
2153 if (argp) {
2154 r = -EFAULT;
2155 if (copy_from_user(&kvm_sigmask, argp,
2156 sizeof kvm_sigmask))
2157 goto out;
2158 r = -EINVAL;
2159 if (kvm_sigmask.len != sizeof csigset)
2160 goto out;
2161 r = -EFAULT;
2162 if (copy_from_user(&csigset, sigmask_arg->sigset,
2163 sizeof csigset))
2164 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002165 sigset_from_compat(&sigset, &csigset);
2166 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2167 } else
2168 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002169 break;
2170 }
2171 default:
2172 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2173 }
2174
2175out:
2176 return r;
2177}
2178#endif
2179
Avi Kivitybccf2152007-02-21 18:04:26 +02002180static long kvm_vm_ioctl(struct file *filp,
2181 unsigned int ioctl, unsigned long arg)
2182{
2183 struct kvm *kvm = filp->private_data;
2184 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002185 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002186
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002187 if (kvm->mm != current->mm)
2188 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002189 switch (ioctl) {
2190 case KVM_CREATE_VCPU:
2191 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002192 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002193 case KVM_SET_USER_MEMORY_REGION: {
2194 struct kvm_userspace_memory_region kvm_userspace_mem;
2195
2196 r = -EFAULT;
2197 if (copy_from_user(&kvm_userspace_mem, argp,
2198 sizeof kvm_userspace_mem))
2199 goto out;
2200
Alex Williamsonf82a8cf2012-12-10 10:33:21 -07002201 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002202 break;
2203 }
2204 case KVM_GET_DIRTY_LOG: {
2205 struct kvm_dirty_log log;
2206
2207 r = -EFAULT;
Al Viro2f3669872007-02-09 16:38:35 +00002208 if (copy_from_user(&log, argp, sizeof log))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002209 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002210 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002211 break;
2212 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002213#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2214 case KVM_REGISTER_COALESCED_MMIO: {
2215 struct kvm_coalesced_mmio_zone zone;
2216 r = -EFAULT;
2217 if (copy_from_user(&zone, argp, sizeof zone))
2218 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002219 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002220 break;
2221 }
2222 case KVM_UNREGISTER_COALESCED_MMIO: {
2223 struct kvm_coalesced_mmio_zone zone;
2224 r = -EFAULT;
2225 if (copy_from_user(&zone, argp, sizeof zone))
2226 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002227 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002228 break;
2229 }
2230#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002231 case KVM_IRQFD: {
2232 struct kvm_irqfd data;
2233
2234 r = -EFAULT;
2235 if (copy_from_user(&data, argp, sizeof data))
2236 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002237 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002238 break;
2239 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002240 case KVM_IOEVENTFD: {
2241 struct kvm_ioeventfd data;
2242
2243 r = -EFAULT;
2244 if (copy_from_user(&data, argp, sizeof data))
2245 goto out;
2246 r = kvm_ioeventfd(kvm, &data);
2247 break;
2248 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002249#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2250 case KVM_SET_BOOT_CPU_ID:
2251 r = 0;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002252 mutex_lock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002253 if (atomic_read(&kvm->online_vcpus) != 0)
2254 r = -EBUSY;
2255 else
2256 kvm->bsp_vcpu_id = arg;
Marcelo Tosatti894a9c52009-06-23 15:05:14 -03002257 mutex_unlock(&kvm->lock);
Gleb Natapov73880c82009-06-09 15:56:28 +03002258 break;
2259#endif
Jan Kiszka07975ad2012-03-29 21:14:12 +02002260#ifdef CONFIG_HAVE_KVM_MSI
2261 case KVM_SIGNAL_MSI: {
2262 struct kvm_msi msi;
2263
2264 r = -EFAULT;
2265 if (copy_from_user(&msi, argp, sizeof msi))
2266 goto out;
2267 r = kvm_send_userspace_msi(kvm, &msi);
2268 break;
2269 }
2270#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002271#ifdef __KVM_HAVE_IRQ_LINE
2272 case KVM_IRQ_LINE_STATUS:
2273 case KVM_IRQ_LINE: {
2274 struct kvm_irq_level irq_event;
2275
2276 r = -EFAULT;
2277 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2278 goto out;
2279
2280 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2281 if (r)
2282 goto out;
2283
2284 r = -EFAULT;
2285 if (ioctl == KVM_IRQ_LINE_STATUS) {
2286 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2287 goto out;
2288 }
2289
2290 r = 0;
2291 break;
2292 }
2293#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002294 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002295 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivitybfd99ff2009-08-26 14:57:50 +03002296 if (r == -ENOTTY)
2297 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002298 }
2299out:
2300 return r;
2301}
2302
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002303#ifdef CONFIG_COMPAT
2304struct compat_kvm_dirty_log {
2305 __u32 slot;
2306 __u32 padding1;
2307 union {
2308 compat_uptr_t dirty_bitmap; /* one bit per page */
2309 __u64 padding2;
2310 };
2311};
2312
2313static long kvm_vm_compat_ioctl(struct file *filp,
2314 unsigned int ioctl, unsigned long arg)
2315{
2316 struct kvm *kvm = filp->private_data;
2317 int r;
2318
2319 if (kvm->mm != current->mm)
2320 return -EIO;
2321 switch (ioctl) {
2322 case KVM_GET_DIRTY_LOG: {
2323 struct compat_kvm_dirty_log compat_log;
2324 struct kvm_dirty_log log;
2325
2326 r = -EFAULT;
2327 if (copy_from_user(&compat_log, (void __user *)arg,
2328 sizeof(compat_log)))
2329 goto out;
2330 log.slot = compat_log.slot;
2331 log.padding1 = compat_log.padding1;
2332 log.padding2 = compat_log.padding2;
2333 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2334
2335 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002336 break;
2337 }
2338 default:
2339 r = kvm_vm_ioctl(filp, ioctl, arg);
2340 }
2341
2342out:
2343 return r;
2344}
2345#endif
2346
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002347static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivityf17abe92007-02-21 19:28:04 +02002348{
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002349 struct page *page[1];
2350 unsigned long addr;
2351 int npages;
2352 gfn_t gfn = vmf->pgoff;
Avi Kivityf17abe92007-02-21 19:28:04 +02002353 struct kvm *kvm = vma->vm_file->private_data;
Avi Kivityf17abe92007-02-21 19:28:04 +02002354
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002355 addr = gfn_to_hva(kvm, gfn);
2356 if (kvm_is_error_hva(addr))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002357 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002358
2359 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2360 NULL);
2361 if (unlikely(npages != 1))
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002362 return VM_FAULT_SIGBUS;
Marcelo Tosatti777b3f42008-09-16 20:54:46 -03002363
2364 vmf->page = page[0];
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002365 return 0;
Avi Kivityf17abe92007-02-21 19:28:04 +02002366}
2367
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002368static const struct vm_operations_struct kvm_vm_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002369 .fault = kvm_vm_fault,
Avi Kivityf17abe92007-02-21 19:28:04 +02002370};
2371
2372static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2373{
2374 vma->vm_ops = &kvm_vm_vm_ops;
2375 return 0;
2376}
2377
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002378static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02002379 .release = kvm_vm_release,
2380 .unlocked_ioctl = kvm_vm_ioctl,
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002381#ifdef CONFIG_COMPAT
2382 .compat_ioctl = kvm_vm_compat_ioctl,
2383#endif
Avi Kivityf17abe92007-02-21 19:28:04 +02002384 .mmap = kvm_vm_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002385 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02002386};
2387
Carsten Ottee08b9632012-01-04 10:25:20 +01002388static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02002389{
Heiko Carstensaac87632010-10-27 17:22:10 +02002390 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002391 struct kvm *kvm;
2392
Carsten Ottee08b9632012-01-04 10:25:20 +01002393 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04002394 if (IS_ERR(kvm))
2395 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09002396#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2397 r = kvm_coalesced_mmio_init(kvm);
2398 if (r < 0) {
2399 kvm_put_kvm(kvm);
2400 return r;
2401 }
2402#endif
Heiko Carstensaac87632010-10-27 17:22:10 +02002403 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2404 if (r < 0)
Al Viro66c0b392008-04-19 20:33:56 +01002405 kvm_put_kvm(kvm);
Avi Kivityf17abe92007-02-21 19:28:04 +02002406
Heiko Carstensaac87632010-10-27 17:22:10 +02002407 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02002408}
2409
Avi Kivity1a811b62008-12-08 18:25:27 +02002410static long kvm_dev_ioctl_check_extension_generic(long arg)
2411{
2412 switch (arg) {
Avi Kivityca9edae2008-12-08 18:29:29 +02002413 case KVM_CAP_USER_MEMORY:
Avi Kivity1a811b62008-12-08 18:25:27 +02002414 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
Jan Kiszka4cd481f2009-04-13 11:59:32 +02002415 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Gleb Natapov73880c82009-06-09 15:56:28 +03002416#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2417 case KVM_CAP_SET_BOOT_CPU_ID:
2418#endif
Avi Kivitya9c73992009-11-04 11:54:59 +02002419 case KVM_CAP_INTERNAL_ERROR_DATA:
Jan Kiszka07975ad2012-03-29 21:14:12 +02002420#ifdef CONFIG_HAVE_KVM_MSI
2421 case KVM_CAP_SIGNAL_MSI:
2422#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002423 return 1;
Marc Zyngier9900b4b2012-06-15 15:07:02 -04002424#ifdef KVM_CAP_IRQ_ROUTING
Avi Kivity399ec802008-11-19 13:58:46 +02002425 case KVM_CAP_IRQ_ROUTING:
Sheng Yang36463142009-03-16 16:33:43 +08002426 return KVM_MAX_IRQ_ROUTES;
Avi Kivity399ec802008-11-19 13:58:46 +02002427#endif
Avi Kivity1a811b62008-12-08 18:25:27 +02002428 default:
2429 break;
2430 }
2431 return kvm_dev_ioctl_check_extension(arg);
2432}
2433
Avi Kivityf17abe92007-02-21 19:28:04 +02002434static long kvm_dev_ioctl(struct file *filp,
2435 unsigned int ioctl, unsigned long arg)
2436{
Avi Kivity07c45a32007-03-07 13:05:38 +02002437 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02002438
2439 switch (ioctl) {
2440 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002441 r = -EINVAL;
2442 if (arg)
2443 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02002444 r = KVM_API_VERSION;
2445 break;
2446 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01002447 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002448 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08002449 case KVM_CHECK_EXTENSION:
Avi Kivity1a811b62008-12-08 18:25:27 +02002450 r = kvm_dev_ioctl_check_extension_generic(arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02002451 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02002452 case KVM_GET_VCPU_MMAP_SIZE:
2453 r = -EINVAL;
2454 if (arg)
2455 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02002456 r = PAGE_SIZE; /* struct kvm_run */
2457#ifdef CONFIG_X86
2458 r += PAGE_SIZE; /* pio data page */
2459#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002460#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2461 r += PAGE_SIZE; /* coalesced mmio ring page */
2462#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02002463 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002464 case KVM_TRACE_ENABLE:
2465 case KVM_TRACE_PAUSE:
2466 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03002467 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04002468 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002469 default:
Carsten Otte043405e2007-10-10 17:16:19 +02002470 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002471 }
2472out:
2473 return r;
2474}
2475
Avi Kivity6aa8b732006-12-10 02:21:36 -08002476static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08002477 .unlocked_ioctl = kvm_dev_ioctl,
2478 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002479 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002480};
2481
2482static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02002483 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002484 "kvm",
2485 &kvm_chardev_ops,
2486};
2487
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002488static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002489{
2490 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02002491 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03002492
Rusty Russell7f59f492008-12-07 21:25:45 +10302493 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002494 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02002495
Rusty Russell7f59f492008-12-07 21:25:45 +10302496 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02002497
2498 r = kvm_arch_hardware_enable(NULL);
2499
2500 if (r) {
2501 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2502 atomic_inc(&hardware_enable_failed);
2503 printk(KERN_INFO "kvm: enabling virtualization on "
2504 "CPU%d failed\n", cpu);
2505 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03002506}
2507
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002508static void hardware_enable(void *junk)
2509{
Jan Kiszkae935b832011-02-08 12:55:33 +01002510 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002511 hardware_enable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002512 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002513}
2514
2515static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03002516{
2517 int cpu = raw_smp_processor_id();
2518
Rusty Russell7f59f492008-12-07 21:25:45 +10302519 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03002520 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10302521 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002522 kvm_arch_hardware_disable(NULL);
Avi Kivity1b6c0162007-05-24 13:03:52 +03002523}
2524
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002525static void hardware_disable(void *junk)
2526{
Jan Kiszkae935b832011-02-08 12:55:33 +01002527 raw_spin_lock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002528 hardware_disable_nolock(junk);
Jan Kiszkae935b832011-02-08 12:55:33 +01002529 raw_spin_unlock(&kvm_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002530}
2531
Alexander Graf10474ae2009-09-15 11:37:46 +02002532static void hardware_disable_all_nolock(void)
2533{
2534 BUG_ON(!kvm_usage_count);
2535
2536 kvm_usage_count--;
2537 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002538 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002539}
2540
2541static void hardware_disable_all(void)
2542{
Jan Kiszkae935b832011-02-08 12:55:33 +01002543 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002544 hardware_disable_all_nolock();
Jan Kiszkae935b832011-02-08 12:55:33 +01002545 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002546}
2547
2548static int hardware_enable_all(void)
2549{
2550 int r = 0;
2551
Jan Kiszkae935b832011-02-08 12:55:33 +01002552 raw_spin_lock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002553
2554 kvm_usage_count++;
2555 if (kvm_usage_count == 1) {
2556 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002557 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02002558
2559 if (atomic_read(&hardware_enable_failed)) {
2560 hardware_disable_all_nolock();
2561 r = -EBUSY;
2562 }
2563 }
2564
Jan Kiszkae935b832011-02-08 12:55:33 +01002565 raw_spin_unlock(&kvm_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02002566
2567 return r;
2568}
2569
Avi Kivity774c47f2007-02-12 00:54:47 -08002570static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2571 void *v)
2572{
2573 int cpu = (long)v;
2574
Alexander Graf10474ae2009-09-15 11:37:46 +02002575 if (!kvm_usage_count)
2576 return NOTIFY_OK;
2577
Avi Kivity1a6f4d72007-11-11 18:37:32 +02002578 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08002579 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03002580 case CPU_DYING:
Avi Kivity6ec8a8562007-08-19 15:57:26 +03002581 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2582 cpu);
2583 hardware_disable(NULL);
2584 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10002585 case CPU_STARTING:
Jeremy Katz43934a32007-02-19 14:37:46 +02002586 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2587 cpu);
Zachary Amsdenda908f22010-08-19 22:07:27 -10002588 hardware_enable(NULL);
Avi Kivity774c47f2007-02-12 00:54:47 -08002589 break;
2590 }
2591 return NOTIFY_OK;
2592}
2593
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002594
Avi Kivityb7c41452010-12-02 17:52:50 +02002595asmlinkage void kvm_spurious_fault(void)
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002596{
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002597 /* Fault while not rebooting. We want the trace. */
2598 BUG();
2599}
Avi Kivityb7c41452010-12-02 17:52:50 +02002600EXPORT_SYMBOL_GPL(kvm_spurious_fault);
Avi Kivity4ecac3f2008-05-13 13:23:38 +03002601
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002602static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04002603 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002604{
Sheng Yang8e1c1812009-04-29 11:09:04 +08002605 /*
2606 * Some (well, at least mine) BIOSes hang on reboot if
2607 * in vmx root mode.
2608 *
2609 * And Intel TXT required VMX off for all cpu when system shutdown.
2610 */
2611 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2612 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002613 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10002614 return NOTIFY_OK;
2615}
2616
2617static struct notifier_block kvm_reboot_notifier = {
2618 .notifier_call = kvm_reboot,
2619 .priority = 0,
2620};
2621
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002622static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002623{
2624 int i;
2625
2626 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03002627 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002628
2629 kvm_iodevice_destructor(pos);
2630 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002631 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002632}
2633
Sasha Levin743eeb02011-07-27 16:00:48 +03002634int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2635{
2636 const struct kvm_io_range *r1 = p1;
2637 const struct kvm_io_range *r2 = p2;
2638
2639 if (r1->addr < r2->addr)
2640 return -1;
2641 if (r1->addr + r1->len > r2->addr + r2->len)
2642 return 1;
2643 return 0;
2644}
2645
2646int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2647 gpa_t addr, int len)
2648{
Sasha Levin743eeb02011-07-27 16:00:48 +03002649 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2650 .addr = addr,
2651 .len = len,
2652 .dev = dev,
2653 };
2654
2655 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2656 kvm_io_bus_sort_cmp, NULL);
2657
2658 return 0;
2659}
2660
2661int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2662 gpa_t addr, int len)
2663{
2664 struct kvm_io_range *range, key;
2665 int off;
2666
2667 key = (struct kvm_io_range) {
2668 .addr = addr,
2669 .len = len,
2670 };
2671
2672 range = bsearch(&key, bus->range, bus->dev_count,
2673 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2674 if (range == NULL)
2675 return -ENOENT;
2676
2677 off = range - bus->range;
2678
2679 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2680 off--;
2681
2682 return off;
2683}
2684
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002685/* kvm_io_bus_write - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002686int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002687 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002688{
Sasha Levin743eeb02011-07-27 16:00:48 +03002689 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002690 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002691 struct kvm_io_range range;
2692
2693 range = (struct kvm_io_range) {
2694 .addr = addr,
2695 .len = len,
2696 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002697
2698 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002699 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2700 if (idx < 0)
2701 return -EOPNOTSUPP;
2702
2703 while (idx < bus->dev_count &&
2704 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2705 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002706 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002707 idx++;
2708 }
2709
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002710 return -EOPNOTSUPP;
2711}
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002712
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002713/* kvm_io_bus_read - called under kvm->slots_lock */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002714int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2715 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002716{
Sasha Levin743eeb02011-07-27 16:00:48 +03002717 int idx;
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002718 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03002719 struct kvm_io_range range;
2720
2721 range = (struct kvm_io_range) {
2722 .addr = addr,
2723 .len = len,
2724 };
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002725
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08002726 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
Sasha Levin743eeb02011-07-27 16:00:48 +03002727 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2728 if (idx < 0)
2729 return -EOPNOTSUPP;
2730
2731 while (idx < bus->dev_count &&
2732 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2733 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002734 return 0;
Sasha Levin743eeb02011-07-27 16:00:48 +03002735 idx++;
2736 }
2737
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03002738 return -EOPNOTSUPP;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002739}
2740
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002741/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03002742int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2743 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002744{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002745 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002746
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002747 bus = kvm->buses[bus_idx];
Amos Konga13007162012-03-09 12:17:32 +08002748 if (bus->dev_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002749 return -ENOSPC;
2750
Amos Konga13007162012-03-09 12:17:32 +08002751 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2752 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002753 if (!new_bus)
2754 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08002755 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2756 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03002757 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002758 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2759 synchronize_srcu_expedited(&kvm->srcu);
2760 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04002761
2762 return 0;
2763}
2764
Marcelo Tosatti79fac952009-12-23 14:35:26 -02002765/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002766int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2767 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04002768{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002769 int i, r;
2770 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03002771
Sasha Levincdfca7b2011-12-04 19:36:28 +02002772 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002773 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08002774 for (i = 0; i < bus->dev_count; i++)
2775 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002776 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04002777 break;
2778 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002779
Amos Konga13007162012-03-09 12:17:32 +08002780 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002781 return r;
Amos Konga13007162012-03-09 12:17:32 +08002782
2783 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2784 sizeof(struct kvm_io_range)), GFP_KERNEL);
2785 if (!new_bus)
2786 return -ENOMEM;
2787
2788 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2789 new_bus->dev_count--;
2790 memcpy(new_bus->range + i, bus->range + i + 1,
2791 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02002792
2793 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2794 synchronize_srcu_expedited(&kvm->srcu);
2795 kfree(bus);
2796 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04002797}
2798
Avi Kivity774c47f2007-02-12 00:54:47 -08002799static struct notifier_block kvm_cpu_notifier = {
2800 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08002801};
2802
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002803static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02002804{
2805 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02002806 struct kvm *kvm;
2807
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002808 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002809 raw_spin_lock(&kvm_lock);
Avi Kivityba1389b2007-11-18 16:24:12 +02002810 list_for_each_entry(kvm, &vm_list, vm_list)
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002811 *val += *(u32 *)((void *)kvm + offset);
Jan Kiszkae935b832011-02-08 12:55:33 +01002812 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002813 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02002814}
2815
2816DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2817
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002818static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03002819{
2820 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002821 struct kvm *kvm;
2822 struct kvm_vcpu *vcpu;
2823 int i;
2824
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002825 *val = 0;
Jan Kiszkae935b832011-02-08 12:55:33 +01002826 raw_spin_lock(&kvm_lock);
Avi Kivity1165f5f2007-04-19 17:27:43 +03002827 list_for_each_entry(kvm, &vm_list, vm_list)
Gleb Natapov988a2ca2009-06-09 15:56:29 +03002828 kvm_for_each_vcpu(i, vcpu, kvm)
2829 *val += *(u32 *)((void *)vcpu + offset);
2830
Jan Kiszkae935b832011-02-08 12:55:33 +01002831 raw_spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08002832 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03002833}
2834
Avi Kivityba1389b2007-11-18 16:24:12 +02002835DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2836
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002837static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02002838 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2839 [KVM_STAT_VM] = &vm_stat_fops,
2840};
Avi Kivity1165f5f2007-04-19 17:27:43 +03002841
Hamo4f69b682011-12-15 14:23:16 +08002842static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002843{
Hamo4f69b682011-12-15 14:23:16 +08002844 int r = -EFAULT;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002845 struct kvm_stats_debugfs_item *p;
2846
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002847 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08002848 if (kvm_debugfs_dir == NULL)
2849 goto out;
2850
2851 for (p = debugfs_entries; p->name; ++p) {
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002852 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
Avi Kivity1165f5f2007-04-19 17:27:43 +03002853 (void *)(long)p->offset,
Avi Kivityba1389b2007-11-18 16:24:12 +02002854 stat_fops[p->kind]);
Hamo4f69b682011-12-15 14:23:16 +08002855 if (p->dentry == NULL)
2856 goto out_dir;
2857 }
2858
2859 return 0;
2860
2861out_dir:
2862 debugfs_remove_recursive(kvm_debugfs_dir);
2863out:
2864 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002865}
2866
2867static void kvm_exit_debug(void)
2868{
2869 struct kvm_stats_debugfs_item *p;
2870
2871 for (p = debugfs_entries; p->name; ++p)
2872 debugfs_remove(p->dentry);
Hollis Blanchard76f7c872008-04-15 16:05:42 -05002873 debugfs_remove(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002874}
2875
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002876static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002877{
Alexander Graf10474ae2009-09-15 11:37:46 +02002878 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002879 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08002880 return 0;
2881}
2882
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002883static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08002884{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002885 if (kvm_usage_count) {
Jan Kiszkae935b832011-02-08 12:55:33 +01002886 WARN_ON(raw_spin_is_locked(&kvm_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09002887 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10002888 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08002889}
2890
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002891static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08002892 .suspend = kvm_suspend,
2893 .resume = kvm_resume,
2894};
2895
Avi Kivity15ad7142007-07-11 18:17:21 +03002896static inline
2897struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2898{
2899 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2900}
2901
2902static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2903{
2904 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2905
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002906 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002907}
2908
2909static void kvm_sched_out(struct preempt_notifier *pn,
2910 struct task_struct *next)
2911{
2912 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2913
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002914 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03002915}
2916
Avi Kivity0ee75be2010-04-28 15:39:01 +03002917int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10002918 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002919{
2920 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002921 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002922
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08002923 r = kvm_arch_init(opaque);
2924 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002925 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08002926
Avi Kivity8437a612009-06-06 14:52:35 -07002927 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10302928 r = -ENOMEM;
2929 goto out_free_0;
2930 }
2931
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002932 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002933 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10302934 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002935
Yang, Sheng002c7f72007-07-31 14:23:01 +03002936 for_each_online_cpu(cpu) {
2937 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002938 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02002939 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03002940 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002941 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03002942 }
2943
Avi Kivity774c47f2007-02-12 00:54:47 -08002944 r = register_cpu_notifier(&kvm_cpu_notifier);
2945 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08002946 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002947 register_reboot_notifier(&kvm_reboot_notifier);
2948
Rusty Russellc16f8622007-07-30 21:12:19 +10002949 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03002950 if (!vcpu_align)
2951 vcpu_align = __alignof__(struct kvm_vcpu);
2952 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08002953 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10002954 if (!kvm_vcpu_cache) {
2955 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002956 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10002957 }
2958
Gleb Natapovaf585b92010-10-14 11:22:46 +02002959 r = kvm_async_pf_init();
2960 if (r)
2961 goto out_free;
2962
Avi Kivity6aa8b732006-12-10 02:21:36 -08002963 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002964 kvm_vm_fops.owner = module;
2965 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002966
2967 r = misc_register(&kvm_dev);
2968 if (r) {
Mike Dayd77c26f2007-10-08 09:02:08 -04002969 printk(KERN_ERR "kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02002970 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002971 }
2972
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01002973 register_syscore_ops(&kvm_syscore_ops);
2974
Avi Kivity15ad7142007-07-11 18:17:21 +03002975 kvm_preempt_ops.sched_in = kvm_sched_in;
2976 kvm_preempt_ops.sched_out = kvm_sched_out;
2977
Hamo4f69b682011-12-15 14:23:16 +08002978 r = kvm_init_debug();
2979 if (r) {
2980 printk(KERN_ERR "kvm: create debugfs files failed\n");
2981 goto out_undebugfs;
2982 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07002983
Avi Kivityc7addb92007-09-16 18:58:32 +02002984 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002985
Hamo4f69b682011-12-15 14:23:16 +08002986out_undebugfs:
2987 unregister_syscore_ops(&kvm_syscore_ops);
Gleb Natapovaf585b92010-10-14 11:22:46 +02002988out_unreg:
2989 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08002990out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10002991 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002992out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08002993 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08002994 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08002995out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08002996out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08002997 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10302998out_free_0a:
2999 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003000out_free_0:
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003001 kvm_arch_exit();
Zhang Xiantaod23087842007-11-29 15:35:39 +08003002out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003003 return r;
3004}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003005EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003006
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003007void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003008{
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003009 kvm_exit_debug();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003010 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003011 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003012 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003013 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003014 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003015 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003016 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003017 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003018 kvm_arch_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303019 free_cpumask_var(cpus_hardware_enabled);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003020}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003021EXPORT_SYMBOL_GPL(kvm_exit);