blob: 8a39dda7a3254677df5996c07c965a2f9f5ef8a8 [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
31static struct kmem_cache *async_pf_cache;
32
33int kvm_async_pf_init(void)
34{
35 async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
36
37 if (!async_pf_cache)
38 return -ENOMEM;
39
40 return 0;
41}
42
43void kvm_async_pf_deinit(void)
44{
45 if (async_pf_cache)
46 kmem_cache_destroy(async_pf_cache);
47 async_pf_cache = NULL;
48}
49
50void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
51{
52 INIT_LIST_HEAD(&vcpu->async_pf.done);
53 INIT_LIST_HEAD(&vcpu->async_pf.queue);
54 spin_lock_init(&vcpu->async_pf.lock);
55}
56
57static void async_pf_execute(struct work_struct *work)
58{
59 struct page *page = NULL;
60 struct kvm_async_pf *apf =
61 container_of(work, struct kvm_async_pf, work);
62 struct mm_struct *mm = apf->mm;
63 struct kvm_vcpu *vcpu = apf->vcpu;
64 unsigned long addr = apf->addr;
65 gva_t gva = apf->gva;
66
67 might_sleep();
68
69 use_mm(mm);
70 down_read(&mm->mmap_sem);
71 get_user_pages(current, mm, addr, 1, 1, 0, &page, NULL);
72 up_read(&mm->mmap_sem);
73 unuse_mm(mm);
74
75 spin_lock(&vcpu->async_pf.lock);
76 list_add_tail(&apf->link, &vcpu->async_pf.done);
77 apf->page = page;
78 apf->done = true;
79 spin_unlock(&vcpu->async_pf.lock);
80
81 /*
82 * apf may be freed by kvm_check_async_pf_completion() after
83 * this point
84 */
85
86 trace_kvm_async_pf_completed(addr, page, gva);
87
88 if (waitqueue_active(&vcpu->wq))
89 wake_up_interruptible(&vcpu->wq);
90
91 mmdrop(mm);
92 kvm_put_kvm(vcpu->kvm);
93}
94
95void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
96{
97 /* cancel outstanding work queue item */
98 while (!list_empty(&vcpu->async_pf.queue)) {
99 struct kvm_async_pf *work =
100 list_entry(vcpu->async_pf.queue.next,
101 typeof(*work), queue);
102 cancel_work_sync(&work->work);
103 list_del(&work->queue);
Radim Krčmář28b441e2013-09-04 22:32:23 +0200104 if (!work->done) { /* work was canceled */
105 mmdrop(work->mm);
106 kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
Gleb Natapovaf585b92010-10-14 11:22:46 +0200107 kmem_cache_free(async_pf_cache, work);
Radim Krčmář28b441e2013-09-04 22:32:23 +0200108 }
Gleb Natapovaf585b92010-10-14 11:22:46 +0200109 }
110
111 spin_lock(&vcpu->async_pf.lock);
112 while (!list_empty(&vcpu->async_pf.done)) {
113 struct kvm_async_pf *work =
114 list_entry(vcpu->async_pf.done.next,
115 typeof(*work), link);
116 list_del(&work->link);
Xiao Guangrong32cad842012-08-03 15:42:52 +0800117 if (!is_error_page(work->page))
Xiao Guangrong2b4b5af2012-07-26 11:58:17 +0800118 kvm_release_page_clean(work->page);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200119 kmem_cache_free(async_pf_cache, work);
120 }
121 spin_unlock(&vcpu->async_pf.lock);
122
123 vcpu->async_pf.queued = 0;
124}
125
126void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
127{
128 struct kvm_async_pf *work;
129
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800130 while (!list_empty_careful(&vcpu->async_pf.done) &&
131 kvm_arch_can_inject_async_page_present(vcpu)) {
132 spin_lock(&vcpu->async_pf.lock);
133 work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
134 link);
135 list_del(&work->link);
136 spin_unlock(&vcpu->async_pf.lock);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200137
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800138 if (work->page)
139 kvm_arch_async_page_ready(vcpu, work);
140 kvm_arch_async_page_present(vcpu, work);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200141
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800142 list_del(&work->queue);
143 vcpu->async_pf.queued--;
Xiao Guangrong32cad842012-08-03 15:42:52 +0800144 if (!is_error_page(work->page))
Xiao Guangrong2b4b5af2012-07-26 11:58:17 +0800145 kvm_release_page_clean(work->page);
Xiao Guangrong15096ff2010-11-02 17:35:35 +0800146 kmem_cache_free(async_pf_cache, work);
147 }
Gleb Natapovaf585b92010-10-14 11:22:46 +0200148}
149
150int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
151 struct kvm_arch_async_pf *arch)
152{
153 struct kvm_async_pf *work;
154
155 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
156 return 0;
157
158 /* setup delayed work */
159
160 /*
161 * do alloc nowait since if we are going to sleep anyway we
162 * may as well sleep faulting in page
163 */
164 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT);
165 if (!work)
166 return 0;
167
168 work->page = NULL;
169 work->done = false;
170 work->vcpu = vcpu;
171 work->gva = gva;
172 work->addr = gfn_to_hva(vcpu->kvm, gfn);
173 work->arch = *arch;
174 work->mm = current->mm;
175 atomic_inc(&work->mm->mm_count);
176 kvm_get_kvm(work->vcpu->kvm);
177
178 /* this can't really happen otherwise gfn_to_pfn_async
179 would succeed */
180 if (unlikely(kvm_is_error_hva(work->addr)))
181 goto retry_sync;
182
183 INIT_WORK(&work->work, async_pf_execute);
184 if (!schedule_work(&work->work))
185 goto retry_sync;
186
187 list_add_tail(&work->queue, &vcpu->async_pf.queue);
188 vcpu->async_pf.queued++;
189 kvm_arch_async_page_not_present(vcpu, work);
190 return 1;
191retry_sync:
192 kvm_put_kvm(work->vcpu->kvm);
193 mmdrop(work->mm);
194 kmem_cache_free(async_pf_cache, work);
195 return 0;
196}
Gleb Natapov344d9582010-10-14 11:22:50 +0200197
198int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
199{
200 struct kvm_async_pf *work;
201
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800202 if (!list_empty_careful(&vcpu->async_pf.done))
Gleb Natapov344d9582010-10-14 11:22:50 +0200203 return 0;
204
205 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
206 if (!work)
207 return -ENOMEM;
208
Xiao Guangrong6cede2e2012-08-03 15:41:22 +0800209 work->page = KVM_ERR_PTR_BAD_PAGE;
Gleb Natapov344d9582010-10-14 11:22:50 +0200210 INIT_LIST_HEAD(&work->queue); /* for list_del to work */
211
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800212 spin_lock(&vcpu->async_pf.lock);
Gleb Natapov344d9582010-10-14 11:22:50 +0200213 list_add_tail(&work->link, &vcpu->async_pf.done);
Xiao Guangrong64f638c2010-11-01 17:03:44 +0800214 spin_unlock(&vcpu->async_pf.lock);
215
Gleb Natapov344d9582010-10-14 11:22:50 +0200216 vcpu->async_pf.queued++;
217 return 0;
218}