xref: /openbmc/linux/virt/kvm/async_pf.c (revision 2596e07a)
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 
80 	might_sleep();
81 
82 	get_user_pages_unlocked(NULL, mm, addr, 1, 1, 0, NULL);
83 	kvm_async_page_present_sync(vcpu, apf);
84 
85 	spin_lock(&vcpu->async_pf.lock);
86 	list_add_tail(&apf->link, &vcpu->async_pf.done);
87 	spin_unlock(&vcpu->async_pf.lock);
88 
89 	/*
90 	 * apf may be freed by kvm_check_async_pf_completion() after
91 	 * this point
92 	 */
93 
94 	trace_kvm_async_pf_completed(addr, gva);
95 
96 	/*
97 	 * This memory barrier pairs with prepare_to_wait's set_current_state()
98 	 */
99 	smp_mb();
100 	if (waitqueue_active(&vcpu->wq))
101 		wake_up_interruptible(&vcpu->wq);
102 
103 	mmput(mm);
104 	kvm_put_kvm(vcpu->kvm);
105 }
106 
107 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
108 {
109 	/* cancel outstanding work queue item */
110 	while (!list_empty(&vcpu->async_pf.queue)) {
111 		struct kvm_async_pf *work =
112 			list_entry(vcpu->async_pf.queue.next,
113 				   typeof(*work), queue);
114 		list_del(&work->queue);
115 
116 #ifdef CONFIG_KVM_ASYNC_PF_SYNC
117 		flush_work(&work->work);
118 #else
119 		if (cancel_work_sync(&work->work)) {
120 			mmput(work->mm);
121 			kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
122 			kmem_cache_free(async_pf_cache, work);
123 		}
124 #endif
125 	}
126 
127 	spin_lock(&vcpu->async_pf.lock);
128 	while (!list_empty(&vcpu->async_pf.done)) {
129 		struct kvm_async_pf *work =
130 			list_entry(vcpu->async_pf.done.next,
131 				   typeof(*work), link);
132 		list_del(&work->link);
133 		kmem_cache_free(async_pf_cache, work);
134 	}
135 	spin_unlock(&vcpu->async_pf.lock);
136 
137 	vcpu->async_pf.queued = 0;
138 }
139 
140 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
141 {
142 	struct kvm_async_pf *work;
143 
144 	while (!list_empty_careful(&vcpu->async_pf.done) &&
145 	      kvm_arch_can_inject_async_page_present(vcpu)) {
146 		spin_lock(&vcpu->async_pf.lock);
147 		work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
148 					      link);
149 		list_del(&work->link);
150 		spin_unlock(&vcpu->async_pf.lock);
151 
152 		kvm_arch_async_page_ready(vcpu, work);
153 		kvm_async_page_present_async(vcpu, work);
154 
155 		list_del(&work->queue);
156 		vcpu->async_pf.queued--;
157 		kmem_cache_free(async_pf_cache, work);
158 	}
159 }
160 
161 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
162 		       struct kvm_arch_async_pf *arch)
163 {
164 	struct kvm_async_pf *work;
165 
166 	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
167 		return 0;
168 
169 	/* setup delayed work */
170 
171 	/*
172 	 * do alloc nowait since if we are going to sleep anyway we
173 	 * may as well sleep faulting in page
174 	 */
175 	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
176 	if (!work)
177 		return 0;
178 
179 	work->wakeup_all = false;
180 	work->vcpu = vcpu;
181 	work->gva = gva;
182 	work->addr = hva;
183 	work->arch = *arch;
184 	work->mm = current->mm;
185 	atomic_inc(&work->mm->mm_users);
186 	kvm_get_kvm(work->vcpu->kvm);
187 
188 	/* this can't really happen otherwise gfn_to_pfn_async
189 	   would succeed */
190 	if (unlikely(kvm_is_error_hva(work->addr)))
191 		goto retry_sync;
192 
193 	INIT_WORK(&work->work, async_pf_execute);
194 	if (!schedule_work(&work->work))
195 		goto retry_sync;
196 
197 	list_add_tail(&work->queue, &vcpu->async_pf.queue);
198 	vcpu->async_pf.queued++;
199 	kvm_arch_async_page_not_present(vcpu, work);
200 	return 1;
201 retry_sync:
202 	kvm_put_kvm(work->vcpu->kvm);
203 	mmput(work->mm);
204 	kmem_cache_free(async_pf_cache, work);
205 	return 0;
206 }
207 
208 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
209 {
210 	struct kvm_async_pf *work;
211 
212 	if (!list_empty_careful(&vcpu->async_pf.done))
213 		return 0;
214 
215 	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
216 	if (!work)
217 		return -ENOMEM;
218 
219 	work->wakeup_all = true;
220 	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
221 
222 	spin_lock(&vcpu->async_pf.lock);
223 	list_add_tail(&work->link, &vcpu->async_pf.done);
224 	spin_unlock(&vcpu->async_pf.lock);
225 
226 	vcpu->async_pf.queued++;
227 	return 0;
228 }
229