blob: db9668869f6ff6866a72f278def5770d71190994 [file] [log] [blame]
Gleb Natapovaf585b92010-10-14 11:22:46 +02001/*
2 * kvm asynchronous fault support
3 *
4 * Copyright 2010 Red Hat, Inc.
5 *
6 * Author:
7 * Gleb Natapov <gleb@redhat.com>
8 *
9 * This file is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License
11 * as published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software Foundation,
20 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
21 */
22
23#include <linux/kvm_host.h>
24#include <linux/slab.h>
25#include <linux/module.h>
26#include <linux/mmu_context.h>
27
28#include "async_pf.h"
29#include <trace/events/kvm.h>
30
Dominik Dingele0ead412013-06-06 15:32:37 +020031static inline void kvm_async_page_present_sync(struct kvm_vcpu *vcpu,
32 struct kvm_async_pf *work)
33{
34#ifdef CONFIG_KVM_ASYNC_PF_SYNC
35 kvm_arch_async_page_present(vcpu, work);
36#endif
37}
38static inline void kvm_async_page_present_async(struct kvm_vcpu *vcpu,
39 struct kvm_async_pf *work)
40{
41#ifndef CONFIG_KVM_ASYNC_PF_SYNC
42 kvm_arch_async_page_present(vcpu, work);
43#endif
44}
45
Gleb Natapovaf585b92010-10-14 11:22:46 +020046static struct kmem_cache *async_pf_cache;
47
48int kvm_async_pf_init(void)
49{
50 async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
51
52 if (!async_pf_cache)
53 return -ENOMEM;
54
55 return 0;
56}
57
58void kvm_async_pf_deinit(void)
59{
Markus Elfring4f526962015-11-15 10:40:36 +010060 kmem_cache_destroy(async_pf_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +020061 async_pf_cache = NULL;
62}
63
64void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
65{
66 INIT_LIST_HEAD(&vcpu->async_pf.done);
67 INIT_LIST_HEAD(&vcpu->async_pf.queue);
68 spin_lock_init(&vcpu->async_pf.lock);
69}
70
71static void async_pf_execute(struct work_struct *work)
72{
Gleb Natapovaf585b92010-10-14 11:22:46 +020073 struct kvm_async_pf *apf =
74 container_of(work, struct kvm_async_pf, work);
75 struct mm_struct *mm = apf->mm;
76 struct kvm_vcpu *vcpu = apf->vcpu;
77 unsigned long addr = apf->addr;
78 gva_t gva = apf->gva;
79
80 might_sleep();
81
Dave Hansen1e987792016-02-12 13:01:54 -080082 /*
83 * This work is run asynchromously to the task which owns
84 * mm and might be done in another context, so we must
85 * use FOLL_REMOTE.
86 */
87 __get_user_pages_unlocked(NULL, mm, addr, 1, 1, 0, NULL, FOLL_REMOTE);
88
Dominik Dingele0ead412013-06-06 15:32:37 +020089 kvm_async_page_present_sync(vcpu, apf);
Gleb Natapovaf585b92010-10-14 11:22:46 +020090
91 spin_lock(&vcpu->async_pf.lock);
92 list_add_tail(&apf->link, &vcpu->async_pf.done);
Gleb Natapovaf585b92010-10-14 11:22:46 +020093 spin_unlock(&vcpu->async_pf.lock);
94
95 /*
96 * apf may be freed by kvm_check_async_pf_completion() after
97 * this point
98 */
99
chai wenf2e10662013-10-14 22:22:33 +0800100 trace_kvm_async_pf_completed(addr, gva);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200101
Kosuke Tatsukawa6003a422015-10-09 12:21:55 +0000102 /*
103 * This memory barrier pairs with prepare_to_wait's set_current_state()
104 */
105 smp_mb();
Marcelo Tosatti85773702016-02-19 09:46:39 +0100106 if (swait_active(&vcpu->wq))
107 swake_up(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200108
Oleg Nesterov41c22f62014-04-21 15:26:01 +0200109 mmput(mm);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200110 kvm_put_kvm(vcpu->kvm);
111}
112
113void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
114{
115 /* cancel outstanding work queue item */
116 while (!list_empty(&vcpu->async_pf.queue)) {
117 struct kvm_async_pf *work =
Geliang Tang433da862016-01-01 19:47:15 +0800118 list_first_entry(&vcpu->async_pf.queue,
119 typeof(*work), queue);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200120 list_del(&work->queue);
Dominik Dingel9f2ceda2013-09-03 12:31:16 +0200121
122#ifdef CONFIG_KVM_ASYNC_PF_SYNC
123 flush_work(&work->work);
124#else
Radim Krčmář98fda162013-09-04 22:32:24 +0200125 if (cancel_work_sync(&work->work)) {
Oleg Nesterov41c22f62014-04-21 15:26:01 +0200126 mmput(work->mm);
Radim Krčmář28b441e2013-09-04 22:32:23 +0200127 kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
Gleb Natapovaf585b92010-10-14 11:22:46 +0200128 kmem_cache_free(async_pf_cache, work);
Radim Krčmář28b441e2013-09-04 22:32:23 +0200129 }
Dominik Dingel9f2ceda2013-09-03 12:31:16 +0200130#endif
Gleb Natapovaf585b92010-10-14 11:22:46 +0200131 }
132
133 spin_lock(&vcpu->async_pf.lock);
134 while (!list_empty(&vcpu->async_pf.done)) {
135 struct kvm_async_pf *work =
Geliang Tang433da862016-01-01 19:47:15 +0800136 list_first_entry(&vcpu->async_pf.done,
137 typeof(*work), link);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200138 list_del(&work->link);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200139 kmem_cache_free(async_pf_cache, work);
140 }
141 spin_unlock(&vcpu->async_pf.lock);
142
143 vcpu->async_pf.queued = 0;
144}
145
146void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
147{
148 struct kvm_async_pf *work;
149
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800150 while (!list_empty_careful(&vcpu->async_pf.done) &&
151 kvm_arch_can_inject_async_page_present(vcpu)) {
152 spin_lock(&vcpu->async_pf.lock);
153 work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
154 link);
155 list_del(&work->link);
156 spin_unlock(&vcpu->async_pf.lock);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200157
chai wenf2e10662013-10-14 22:22:33 +0800158 kvm_arch_async_page_ready(vcpu, work);
Dominik Dingel1179ba52014-01-31 14:32:46 +0100159 kvm_async_page_present_async(vcpu, work);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200160
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800161 list_del(&work->queue);
162 vcpu->async_pf.queued--;
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800163 kmem_cache_free(async_pf_cache, work);
164 }
Gleb Natapovaf585b92010-10-14 11:22:46 +0200165}
166
Dominik Dingele0ead412013-06-06 15:32:37 +0200167int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
Gleb Natapovaf585b92010-10-14 11:22:46 +0200168 struct kvm_arch_async_pf *arch)
169{
170 struct kvm_async_pf *work;
171
172 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
173 return 0;
174
175 /* setup delayed work */
176
177 /*
178 * do alloc nowait since if we are going to sleep anyway we
179 * may as well sleep faulting in page
180 */
Christian Borntraegerd7444792016-02-19 13:11:46 +0100181 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200182 if (!work)
183 return 0;
184
chai wenf2e10662013-10-14 22:22:33 +0800185 work->wakeup_all = false;
Gleb Natapovaf585b92010-10-14 11:22:46 +0200186 work->vcpu = vcpu;
187 work->gva = gva;
Dominik Dingele0ead412013-06-06 15:32:37 +0200188 work->addr = hva;
Gleb Natapovaf585b92010-10-14 11:22:46 +0200189 work->arch = *arch;
190 work->mm = current->mm;
Oleg Nesterov41c22f62014-04-21 15:26:01 +0200191 atomic_inc(&work->mm->mm_users);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200192 kvm_get_kvm(work->vcpu->kvm);
193
194 /* this can't really happen otherwise gfn_to_pfn_async
195 would succeed */
196 if (unlikely(kvm_is_error_hva(work->addr)))
197 goto retry_sync;
198
199 INIT_WORK(&work->work, async_pf_execute);
200 if (!schedule_work(&work->work))
201 goto retry_sync;
202
203 list_add_tail(&work->queue, &vcpu->async_pf.queue);
204 vcpu->async_pf.queued++;
205 kvm_arch_async_page_not_present(vcpu, work);
206 return 1;
207retry_sync:
208 kvm_put_kvm(work->vcpu->kvm);
Oleg Nesterov41c22f62014-04-21 15:26:01 +0200209 mmput(work->mm);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200210 kmem_cache_free(async_pf_cache, work);
211 return 0;
212}
Gleb Natapov344d9582010-10-14 11:22:50 +0200213
214int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
215{
216 struct kvm_async_pf *work;
217
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800218 if (!list_empty_careful(&vcpu->async_pf.done))
Gleb Natapov344d9582010-10-14 11:22:50 +0200219 return 0;
220
221 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
222 if (!work)
223 return -ENOMEM;
224
chai wenf2e10662013-10-14 22:22:33 +0800225 work->wakeup_all = true;
Gleb Natapov344d9582010-10-14 11:22:50 +0200226 INIT_LIST_HEAD(&work->queue); /* for list_del to work */
227
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800228 spin_lock(&vcpu->async_pf.lock);
Gleb Natapov344d9582010-10-14 11:22:50 +0200229 list_add_tail(&work->link, &vcpu->async_pf.done);
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800230 spin_unlock(&vcpu->async_pf.lock);
231
Gleb Natapov344d9582010-10-14 11:22:50 +0200232 vcpu->async_pf.queued++;
233 return 0;
234}