xref: /openbmc/linux/virt/kvm/async_pf.c (revision bdd303cb)
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 #include <linux/sched/mm.h>
28 
29 #include "async_pf.h"
30 #include <trace/events/kvm.h>
31 
32 static inline void kvm_async_page_present_sync(struct kvm_vcpu *vcpu,
33 					       struct kvm_async_pf *work)
34 {
35 #ifdef CONFIG_KVM_ASYNC_PF_SYNC
36 	kvm_arch_async_page_present(vcpu, work);
37 #endif
38 }
39 static inline void kvm_async_page_present_async(struct kvm_vcpu *vcpu,
40 						struct kvm_async_pf *work)
41 {
42 #ifndef CONFIG_KVM_ASYNC_PF_SYNC
43 	kvm_arch_async_page_present(vcpu, work);
44 #endif
45 }
46 
47 static struct kmem_cache *async_pf_cache;
48 
49 int kvm_async_pf_init(void)
50 {
51 	async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
52 
53 	if (!async_pf_cache)
54 		return -ENOMEM;
55 
56 	return 0;
57 }
58 
59 void kvm_async_pf_deinit(void)
60 {
61 	kmem_cache_destroy(async_pf_cache);
62 	async_pf_cache = NULL;
63 }
64 
65 void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
66 {
67 	INIT_LIST_HEAD(&vcpu->async_pf.done);
68 	INIT_LIST_HEAD(&vcpu->async_pf.queue);
69 	spin_lock_init(&vcpu->async_pf.lock);
70 }
71 
72 static void async_pf_execute(struct work_struct *work)
73 {
74 	struct kvm_async_pf *apf =
75 		container_of(work, struct kvm_async_pf, work);
76 	struct mm_struct *mm = apf->mm;
77 	struct kvm_vcpu *vcpu = apf->vcpu;
78 	unsigned long addr = apf->addr;
79 	gva_t gva = apf->gva;
80 	int locked = 1;
81 
82 	might_sleep();
83 
84 	/*
85 	 * This work is run asynchronously to the task which owns
86 	 * mm and might be done in another context, so we must
87 	 * access remotely.
88 	 */
89 	down_read(&mm->mmap_sem);
90 	get_user_pages_remote(NULL, mm, addr, 1, FOLL_WRITE, NULL, NULL,
91 			&locked);
92 	if (locked)
93 		up_read(&mm->mmap_sem);
94 
95 	kvm_async_page_present_sync(vcpu, apf);
96 
97 	spin_lock(&vcpu->async_pf.lock);
98 	list_add_tail(&apf->link, &vcpu->async_pf.done);
99 	apf->vcpu = NULL;
100 	spin_unlock(&vcpu->async_pf.lock);
101 
102 	/*
103 	 * apf may be freed by kvm_check_async_pf_completion() after
104 	 * this point
105 	 */
106 
107 	trace_kvm_async_pf_completed(addr, gva);
108 
109 	if (swq_has_sleeper(&vcpu->wq))
110 		swake_up_one(&vcpu->wq);
111 
112 	mmput(mm);
113 	kvm_put_kvm(vcpu->kvm);
114 }
115 
116 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
117 {
118 	spin_lock(&vcpu->async_pf.lock);
119 
120 	/* cancel outstanding work queue item */
121 	while (!list_empty(&vcpu->async_pf.queue)) {
122 		struct kvm_async_pf *work =
123 			list_first_entry(&vcpu->async_pf.queue,
124 					 typeof(*work), queue);
125 		list_del(&work->queue);
126 
127 		/*
128 		 * We know it's present in vcpu->async_pf.done, do
129 		 * nothing here.
130 		 */
131 		if (!work->vcpu)
132 			continue;
133 
134 		spin_unlock(&vcpu->async_pf.lock);
135 #ifdef CONFIG_KVM_ASYNC_PF_SYNC
136 		flush_work(&work->work);
137 #else
138 		if (cancel_work_sync(&work->work)) {
139 			mmput(work->mm);
140 			kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
141 			kmem_cache_free(async_pf_cache, work);
142 		}
143 #endif
144 		spin_lock(&vcpu->async_pf.lock);
145 	}
146 
147 	while (!list_empty(&vcpu->async_pf.done)) {
148 		struct kvm_async_pf *work =
149 			list_first_entry(&vcpu->async_pf.done,
150 					 typeof(*work), link);
151 		list_del(&work->link);
152 		kmem_cache_free(async_pf_cache, work);
153 	}
154 	spin_unlock(&vcpu->async_pf.lock);
155 
156 	vcpu->async_pf.queued = 0;
157 }
158 
159 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
160 {
161 	struct kvm_async_pf *work;
162 
163 	while (!list_empty_careful(&vcpu->async_pf.done) &&
164 	      kvm_arch_can_inject_async_page_present(vcpu)) {
165 		spin_lock(&vcpu->async_pf.lock);
166 		work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
167 					      link);
168 		list_del(&work->link);
169 		spin_unlock(&vcpu->async_pf.lock);
170 
171 		kvm_arch_async_page_ready(vcpu, work);
172 		kvm_async_page_present_async(vcpu, work);
173 
174 		list_del(&work->queue);
175 		vcpu->async_pf.queued--;
176 		kmem_cache_free(async_pf_cache, work);
177 	}
178 }
179 
180 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
181 		       struct kvm_arch_async_pf *arch)
182 {
183 	struct kvm_async_pf *work;
184 
185 	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
186 		return 0;
187 
188 	/* setup delayed work */
189 
190 	/*
191 	 * do alloc nowait since if we are going to sleep anyway we
192 	 * may as well sleep faulting in page
193 	 */
194 	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
195 	if (!work)
196 		return 0;
197 
198 	work->wakeup_all = false;
199 	work->vcpu = vcpu;
200 	work->gva = gva;
201 	work->addr = hva;
202 	work->arch = *arch;
203 	work->mm = current->mm;
204 	mmget(work->mm);
205 	kvm_get_kvm(work->vcpu->kvm);
206 
207 	/* this can't really happen otherwise gfn_to_pfn_async
208 	   would succeed */
209 	if (unlikely(kvm_is_error_hva(work->addr)))
210 		goto retry_sync;
211 
212 	INIT_WORK(&work->work, async_pf_execute);
213 	if (!schedule_work(&work->work))
214 		goto retry_sync;
215 
216 	list_add_tail(&work->queue, &vcpu->async_pf.queue);
217 	vcpu->async_pf.queued++;
218 	kvm_arch_async_page_not_present(vcpu, work);
219 	return 1;
220 retry_sync:
221 	kvm_put_kvm(work->vcpu->kvm);
222 	mmput(work->mm);
223 	kmem_cache_free(async_pf_cache, work);
224 	return 0;
225 }
226 
227 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
228 {
229 	struct kvm_async_pf *work;
230 
231 	if (!list_empty_careful(&vcpu->async_pf.done))
232 		return 0;
233 
234 	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
235 	if (!work)
236 		return -ENOMEM;
237 
238 	work->wakeup_all = true;
239 	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
240 
241 	spin_lock(&vcpu->async_pf.lock);
242 	list_add_tail(&work->link, &vcpu->async_pf.done);
243 	spin_unlock(&vcpu->async_pf.lock);
244 
245 	vcpu->async_pf.queued++;
246 	return 0;
247 }
248