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 struct kmem_cache *async_pf_cache;
32 
kvm_async_pf_init(void)33 int 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 
kvm_async_pf_deinit(void)43 void 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 
kvm_async_pf_vcpu_init(struct kvm_vcpu * vcpu)50 void 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 
async_pf_execute(struct work_struct * work)57 static 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 
kvm_clear_async_pf_completion_queue(struct kvm_vcpu * vcpu)95 void 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);
104 		if (!work->done) /* work was canceled */
105 			kmem_cache_free(async_pf_cache, work);
106 	}
107 
108 	spin_lock(&vcpu->async_pf.lock);
109 	while (!list_empty(&vcpu->async_pf.done)) {
110 		struct kvm_async_pf *work =
111 			list_entry(vcpu->async_pf.done.next,
112 				   typeof(*work), link);
113 		list_del(&work->link);
114 		if (work->page)
115 			put_page(work->page);
116 		kmem_cache_free(async_pf_cache, work);
117 	}
118 	spin_unlock(&vcpu->async_pf.lock);
119 
120 	vcpu->async_pf.queued = 0;
121 }
122 
kvm_check_async_pf_completion(struct kvm_vcpu * vcpu)123 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
124 {
125 	struct kvm_async_pf *work;
126 
127 	while (!list_empty_careful(&vcpu->async_pf.done) &&
128 	      kvm_arch_can_inject_async_page_present(vcpu)) {
129 		spin_lock(&vcpu->async_pf.lock);
130 		work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
131 					      link);
132 		list_del(&work->link);
133 		spin_unlock(&vcpu->async_pf.lock);
134 
135 		if (work->page)
136 			kvm_arch_async_page_ready(vcpu, work);
137 		kvm_arch_async_page_present(vcpu, work);
138 
139 		list_del(&work->queue);
140 		vcpu->async_pf.queued--;
141 		if (work->page)
142 			put_page(work->page);
143 		kmem_cache_free(async_pf_cache, work);
144 	}
145 }
146 
kvm_setup_async_pf(struct kvm_vcpu * vcpu,gva_t gva,gfn_t gfn,struct kvm_arch_async_pf * arch)147 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
148 		       struct kvm_arch_async_pf *arch)
149 {
150 	struct kvm_async_pf *work;
151 
152 	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
153 		return 0;
154 
155 	/* setup delayed work */
156 
157 	/*
158 	 * do alloc nowait since if we are going to sleep anyway we
159 	 * may as well sleep faulting in page
160 	 */
161 	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT);
162 	if (!work)
163 		return 0;
164 
165 	work->page = NULL;
166 	work->done = false;
167 	work->vcpu = vcpu;
168 	work->gva = gva;
169 	work->addr = gfn_to_hva(vcpu->kvm, gfn);
170 	work->arch = *arch;
171 	work->mm = current->mm;
172 	atomic_inc(&work->mm->mm_count);
173 	kvm_get_kvm(work->vcpu->kvm);
174 
175 	/* this can't really happen otherwise gfn_to_pfn_async
176 	   would succeed */
177 	if (unlikely(kvm_is_error_hva(work->addr)))
178 		goto retry_sync;
179 
180 	INIT_WORK(&work->work, async_pf_execute);
181 	if (!schedule_work(&work->work))
182 		goto retry_sync;
183 
184 	list_add_tail(&work->queue, &vcpu->async_pf.queue);
185 	vcpu->async_pf.queued++;
186 	kvm_arch_async_page_not_present(vcpu, work);
187 	return 1;
188 retry_sync:
189 	kvm_put_kvm(work->vcpu->kvm);
190 	mmdrop(work->mm);
191 	kmem_cache_free(async_pf_cache, work);
192 	return 0;
193 }
194 
kvm_async_pf_wakeup_all(struct kvm_vcpu * vcpu)195 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
196 {
197 	struct kvm_async_pf *work;
198 
199 	if (!list_empty_careful(&vcpu->async_pf.done))
200 		return 0;
201 
202 	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
203 	if (!work)
204 		return -ENOMEM;
205 
206 	work->page = bad_page;
207 	get_page(bad_page);
208 	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
209 
210 	spin_lock(&vcpu->async_pf.lock);
211 	list_add_tail(&work->link, &vcpu->async_pf.done);
212 	spin_unlock(&vcpu->async_pf.lock);
213 
214 	vcpu->async_pf.queued++;
215 	return 0;
216 }
217