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
31static 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}
38static 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
46static struct kmem_cache *async_pf_cache;
47
48int 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
58void kvm_async_pf_deinit(void)
59{
60	if (async_pf_cache)
61		kmem_cache_destroy(async_pf_cache);
62	async_pf_cache = NULL;
63}
64
65void 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
72static 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
81	might_sleep();
82
83	get_user_pages_unlocked(NULL, mm, addr, 1, 1, 0, NULL);
84	kvm_async_page_present_sync(vcpu, apf);
85
86	spin_lock(&vcpu->async_pf.lock);
87	list_add_tail(&apf->link, &vcpu->async_pf.done);
88	spin_unlock(&vcpu->async_pf.lock);
89
90	/*
91	 * apf may be freed by kvm_check_async_pf_completion() after
92	 * this point
93	 */
94
95	trace_kvm_async_pf_completed(addr, gva);
96
97	if (waitqueue_active(&vcpu->wq))
98		wake_up_interruptible(&vcpu->wq);
99
100	mmput(mm);
101	kvm_put_kvm(vcpu->kvm);
102}
103
104void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
105{
106	/* cancel outstanding work queue item */
107	while (!list_empty(&vcpu->async_pf.queue)) {
108		struct kvm_async_pf *work =
109			list_entry(vcpu->async_pf.queue.next,
110				   typeof(*work), queue);
111		list_del(&work->queue);
112
113#ifdef CONFIG_KVM_ASYNC_PF_SYNC
114		flush_work(&work->work);
115#else
116		if (cancel_work_sync(&work->work)) {
117			mmput(work->mm);
118			kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
119			kmem_cache_free(async_pf_cache, work);
120		}
121#endif
122	}
123
124	spin_lock(&vcpu->async_pf.lock);
125	while (!list_empty(&vcpu->async_pf.done)) {
126		struct kvm_async_pf *work =
127			list_entry(vcpu->async_pf.done.next,
128				   typeof(*work), link);
129		list_del(&work->link);
130		kmem_cache_free(async_pf_cache, work);
131	}
132	spin_unlock(&vcpu->async_pf.lock);
133
134	vcpu->async_pf.queued = 0;
135}
136
137void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
138{
139	struct kvm_async_pf *work;
140
141	while (!list_empty_careful(&vcpu->async_pf.done) &&
142	      kvm_arch_can_inject_async_page_present(vcpu)) {
143		spin_lock(&vcpu->async_pf.lock);
144		work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
145					      link);
146		list_del(&work->link);
147		spin_unlock(&vcpu->async_pf.lock);
148
149		kvm_arch_async_page_ready(vcpu, work);
150		kvm_async_page_present_async(vcpu, work);
151
152		list_del(&work->queue);
153		vcpu->async_pf.queued--;
154		kmem_cache_free(async_pf_cache, work);
155	}
156}
157
158int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
159		       struct kvm_arch_async_pf *arch)
160{
161	struct kvm_async_pf *work;
162
163	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
164		return 0;
165
166	/* setup delayed work */
167
168	/*
169	 * do alloc nowait since if we are going to sleep anyway we
170	 * may as well sleep faulting in page
171	 */
172	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
173	if (!work)
174		return 0;
175
176	work->wakeup_all = false;
177	work->vcpu = vcpu;
178	work->gva = gva;
179	work->addr = hva;
180	work->arch = *arch;
181	work->mm = current->mm;
182	atomic_inc(&work->mm->mm_users);
183	kvm_get_kvm(work->vcpu->kvm);
184
185	/* this can't really happen otherwise gfn_to_pfn_async
186	   would succeed */
187	if (unlikely(kvm_is_error_hva(work->addr)))
188		goto retry_sync;
189
190	INIT_WORK(&work->work, async_pf_execute);
191	if (!schedule_work(&work->work))
192		goto retry_sync;
193
194	list_add_tail(&work->queue, &vcpu->async_pf.queue);
195	vcpu->async_pf.queued++;
196	kvm_arch_async_page_not_present(vcpu, work);
197	return 1;
198retry_sync:
199	kvm_put_kvm(work->vcpu->kvm);
200	mmput(work->mm);
201	kmem_cache_free(async_pf_cache, work);
202	return 0;
203}
204
205int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
206{
207	struct kvm_async_pf *work;
208
209	if (!list_empty_careful(&vcpu->async_pf.done))
210		return 0;
211
212	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
213	if (!work)
214		return -ENOMEM;
215
216	work->wakeup_all = true;
217	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
218
219	spin_lock(&vcpu->async_pf.lock);
220	list_add_tail(&work->link, &vcpu->async_pf.done);
221	spin_unlock(&vcpu->async_pf.lock);
222
223	vcpu->async_pf.queued++;
224	return 0;
225}
226