]>
Commit | Line | Data |
---|---|---|
1 | /* | |
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 | ||
31 | static 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 | } | |
38 | static 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 | ||
46 | static struct kmem_cache *async_pf_cache; | |
47 | ||
48 | int 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 | ||
58 | void kvm_async_pf_deinit(void) | |
59 | { | |
60 | kmem_cache_destroy(async_pf_cache); | |
61 | async_pf_cache = NULL; | |
62 | } | |
63 | ||
64 | void 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 | ||
71 | static void async_pf_execute(struct work_struct *work) | |
72 | { | |
73 | 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 | int locked = 1; | |
80 | ||
81 | might_sleep(); | |
82 | ||
83 | /* | |
84 | * This work is run asynchromously to the task which owns | |
85 | * mm and might be done in another context, so we must | |
86 | * access remotely. | |
87 | */ | |
88 | down_read(&mm->mmap_sem); | |
89 | get_user_pages_remote(NULL, mm, addr, 1, FOLL_WRITE, NULL, NULL, | |
90 | &locked); | |
91 | if (locked) | |
92 | up_read(&mm->mmap_sem); | |
93 | ||
94 | kvm_async_page_present_sync(vcpu, apf); | |
95 | ||
96 | spin_lock(&vcpu->async_pf.lock); | |
97 | list_add_tail(&apf->link, &vcpu->async_pf.done); | |
98 | apf->vcpu = NULL; | |
99 | spin_unlock(&vcpu->async_pf.lock); | |
100 | ||
101 | /* | |
102 | * apf may be freed by kvm_check_async_pf_completion() after | |
103 | * this point | |
104 | */ | |
105 | ||
106 | trace_kvm_async_pf_completed(addr, gva); | |
107 | ||
108 | /* | |
109 | * This memory barrier pairs with prepare_to_wait's set_current_state() | |
110 | */ | |
111 | smp_mb(); | |
112 | if (swait_active(&vcpu->wq)) | |
113 | swake_up(&vcpu->wq); | |
114 | ||
115 | mmput(mm); | |
116 | kvm_put_kvm(vcpu->kvm); | |
117 | } | |
118 | ||
119 | void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) | |
120 | { | |
121 | spin_lock(&vcpu->async_pf.lock); | |
122 | ||
123 | /* cancel outstanding work queue item */ | |
124 | while (!list_empty(&vcpu->async_pf.queue)) { | |
125 | struct kvm_async_pf *work = | |
126 | list_first_entry(&vcpu->async_pf.queue, | |
127 | typeof(*work), queue); | |
128 | list_del(&work->queue); | |
129 | ||
130 | /* | |
131 | * We know it's present in vcpu->async_pf.done, do | |
132 | * nothing here. | |
133 | */ | |
134 | if (!work->vcpu) | |
135 | continue; | |
136 | ||
137 | spin_unlock(&vcpu->async_pf.lock); | |
138 | #ifdef CONFIG_KVM_ASYNC_PF_SYNC | |
139 | flush_work(&work->work); | |
140 | #else | |
141 | if (cancel_work_sync(&work->work)) { | |
142 | mmput(work->mm); | |
143 | kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */ | |
144 | kmem_cache_free(async_pf_cache, work); | |
145 | } | |
146 | #endif | |
147 | spin_lock(&vcpu->async_pf.lock); | |
148 | } | |
149 | ||
150 | while (!list_empty(&vcpu->async_pf.done)) { | |
151 | struct kvm_async_pf *work = | |
152 | list_first_entry(&vcpu->async_pf.done, | |
153 | typeof(*work), link); | |
154 | list_del(&work->link); | |
155 | kmem_cache_free(async_pf_cache, work); | |
156 | } | |
157 | spin_unlock(&vcpu->async_pf.lock); | |
158 | ||
159 | vcpu->async_pf.queued = 0; | |
160 | } | |
161 | ||
162 | void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu) | |
163 | { | |
164 | struct kvm_async_pf *work; | |
165 | ||
166 | while (!list_empty_careful(&vcpu->async_pf.done) && | |
167 | kvm_arch_can_inject_async_page_present(vcpu)) { | |
168 | spin_lock(&vcpu->async_pf.lock); | |
169 | work = list_first_entry(&vcpu->async_pf.done, typeof(*work), | |
170 | link); | |
171 | list_del(&work->link); | |
172 | spin_unlock(&vcpu->async_pf.lock); | |
173 | ||
174 | kvm_arch_async_page_ready(vcpu, work); | |
175 | kvm_async_page_present_async(vcpu, work); | |
176 | ||
177 | list_del(&work->queue); | |
178 | vcpu->async_pf.queued--; | |
179 | kmem_cache_free(async_pf_cache, work); | |
180 | } | |
181 | } | |
182 | ||
183 | int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva, | |
184 | struct kvm_arch_async_pf *arch) | |
185 | { | |
186 | struct kvm_async_pf *work; | |
187 | ||
188 | if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU) | |
189 | return 0; | |
190 | ||
191 | /* setup delayed work */ | |
192 | ||
193 | /* | |
194 | * do alloc nowait since if we are going to sleep anyway we | |
195 | * may as well sleep faulting in page | |
196 | */ | |
197 | work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN); | |
198 | if (!work) | |
199 | return 0; | |
200 | ||
201 | work->wakeup_all = false; | |
202 | work->vcpu = vcpu; | |
203 | work->gva = gva; | |
204 | work->addr = hva; | |
205 | work->arch = *arch; | |
206 | work->mm = current->mm; | |
207 | atomic_inc(&work->mm->mm_users); | |
208 | kvm_get_kvm(work->vcpu->kvm); | |
209 | ||
210 | /* this can't really happen otherwise gfn_to_pfn_async | |
211 | would succeed */ | |
212 | if (unlikely(kvm_is_error_hva(work->addr))) | |
213 | goto retry_sync; | |
214 | ||
215 | INIT_WORK(&work->work, async_pf_execute); | |
216 | if (!schedule_work(&work->work)) | |
217 | goto retry_sync; | |
218 | ||
219 | list_add_tail(&work->queue, &vcpu->async_pf.queue); | |
220 | vcpu->async_pf.queued++; | |
221 | kvm_arch_async_page_not_present(vcpu, work); | |
222 | return 1; | |
223 | retry_sync: | |
224 | kvm_put_kvm(work->vcpu->kvm); | |
225 | mmput(work->mm); | |
226 | kmem_cache_free(async_pf_cache, work); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu) | |
231 | { | |
232 | struct kvm_async_pf *work; | |
233 | ||
234 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
235 | return 0; | |
236 | ||
237 | work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC); | |
238 | if (!work) | |
239 | return -ENOMEM; | |
240 | ||
241 | work->wakeup_all = true; | |
242 | INIT_LIST_HEAD(&work->queue); /* for list_del to work */ | |
243 | ||
244 | spin_lock(&vcpu->async_pf.lock); | |
245 | list_add_tail(&work->link, &vcpu->async_pf.done); | |
246 | spin_unlock(&vcpu->async_pf.lock); | |
247 | ||
248 | vcpu->async_pf.queued++; | |
249 | return 0; | |
250 | } |