Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kvm asynchronous fault support
4 *
5 * Copyright 2010 Red Hat, Inc.
6 *
7 * Author:
8 * Gleb Natapov <gleb@redhat.com>
9 */
10
11#include <linux/kvm_host.h>
12#include <linux/slab.h>
13#include <linux/module.h>
14#include <linux/mmu_context.h>
15#include <linux/sched/mm.h>
16
17#include "async_pf.h"
18#include <trace/events/kvm.h>
19
20static struct kmem_cache *async_pf_cache;
21
22int kvm_async_pf_init(void)
23{
24 async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
25
26 if (!async_pf_cache)
27 return -ENOMEM;
28
29 return 0;
30}
31
32void kvm_async_pf_deinit(void)
33{
34 kmem_cache_destroy(async_pf_cache);
35 async_pf_cache = NULL;
36}
37
38void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
39{
40 INIT_LIST_HEAD(&vcpu->async_pf.done);
41 INIT_LIST_HEAD(&vcpu->async_pf.queue);
42 spin_lock_init(&vcpu->async_pf.lock);
43}
44
45static void async_pf_execute(struct work_struct *work)
46{
47 struct kvm_async_pf *apf =
48 container_of(work, struct kvm_async_pf, work);
49 struct kvm_vcpu *vcpu = apf->vcpu;
50 struct mm_struct *mm = vcpu->kvm->mm;
51 unsigned long addr = apf->addr;
52 gpa_t cr2_or_gpa = apf->cr2_or_gpa;
53 int locked = 1;
54 bool first;
55
56 might_sleep();
57
58 /*
59 * Attempt to pin the VM's host address space, and simply skip gup() if
60 * acquiring a pin fail, i.e. if the process is exiting. Note, KVM
61 * holds a reference to its associated mm_struct until the very end of
62 * kvm_destroy_vm(), i.e. the struct itself won't be freed before this
63 * work item is fully processed.
64 */
65 if (mmget_not_zero(mm)) {
66 mmap_read_lock(mm);
67 get_user_pages_remote(mm, addr, 1, FOLL_WRITE, NULL, &locked);
68 if (locked)
69 mmap_read_unlock(mm);
70 mmput(mm);
71 }
72
73 /*
74 * Notify and kick the vCPU even if faulting in the page failed, e.g.
75 * so that the vCPU can retry the fault synchronously.
76 */
77 if (IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
78 kvm_arch_async_page_present(vcpu, apf);
79
80 spin_lock(&vcpu->async_pf.lock);
81 first = list_empty(&vcpu->async_pf.done);
82 list_add_tail(&apf->link, &vcpu->async_pf.done);
83 spin_unlock(&vcpu->async_pf.lock);
84
85 /*
86 * The apf struct may be freed by kvm_check_async_pf_completion() as
87 * soon as the lock is dropped. Nullify it to prevent improper usage.
88 */
89 apf = NULL;
90
91 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
92 kvm_arch_async_page_present_queued(vcpu);
93
94 trace_kvm_async_pf_completed(addr, cr2_or_gpa);
95
96 __kvm_vcpu_wake_up(vcpu);
97}
98
99static void kvm_flush_and_free_async_pf_work(struct kvm_async_pf *work)
100{
101 /*
102 * The async #PF is "done", but KVM must wait for the work item itself,
103 * i.e. async_pf_execute(), to run to completion. If KVM is a module,
104 * KVM must ensure *no* code owned by the KVM (the module) can be run
105 * after the last call to module_put(). Note, flushing the work item
106 * is always required when the item is taken off the completion queue.
107 * E.g. even if the vCPU handles the item in the "normal" path, the VM
108 * could be terminated before async_pf_execute() completes.
109 *
110 * Wake all events skip the queue and go straight done, i.e. don't
111 * need to be flushed (but sanity check that the work wasn't queued).
112 */
113 if (work->wakeup_all)
114 WARN_ON_ONCE(work->work.func);
115 else
116 flush_work(&work->work);
117 kmem_cache_free(async_pf_cache, work);
118}
119
120void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
121{
122 /* cancel outstanding work queue item */
123 while (!list_empty(&vcpu->async_pf.queue)) {
124 struct kvm_async_pf *work =
125 list_first_entry(&vcpu->async_pf.queue,
126 typeof(*work), queue);
127 list_del(&work->queue);
128
129#ifdef CONFIG_KVM_ASYNC_PF_SYNC
130 flush_work(&work->work);
131#else
132 if (cancel_work_sync(&work->work))
133 kmem_cache_free(async_pf_cache, work);
134#endif
135 }
136
137 spin_lock(&vcpu->async_pf.lock);
138 while (!list_empty(&vcpu->async_pf.done)) {
139 struct kvm_async_pf *work =
140 list_first_entry(&vcpu->async_pf.done,
141 typeof(*work), link);
142 list_del(&work->link);
143
144 spin_unlock(&vcpu->async_pf.lock);
145 kvm_flush_and_free_async_pf_work(work);
146 spin_lock(&vcpu->async_pf.lock);
147 }
148 spin_unlock(&vcpu->async_pf.lock);
149
150 vcpu->async_pf.queued = 0;
151}
152
153void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
154{
155 struct kvm_async_pf *work;
156
157 while (!list_empty_careful(&vcpu->async_pf.done) &&
158 kvm_arch_can_dequeue_async_page_present(vcpu)) {
159 spin_lock(&vcpu->async_pf.lock);
160 work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
161 link);
162 list_del(&work->link);
163 spin_unlock(&vcpu->async_pf.lock);
164
165 kvm_arch_async_page_ready(vcpu, work);
166 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
167 kvm_arch_async_page_present(vcpu, work);
168
169 list_del(&work->queue);
170 vcpu->async_pf.queued--;
171 kvm_flush_and_free_async_pf_work(work);
172 }
173}
174
175/*
176 * Try to schedule a job to handle page fault asynchronously. Returns 'true' on
177 * success, 'false' on failure (page fault has to be handled synchronously).
178 */
179bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
180 unsigned long hva, struct kvm_arch_async_pf *arch)
181{
182 struct kvm_async_pf *work;
183
184 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
185 return false;
186
187 /* Arch specific code should not do async PF in this case */
188 if (unlikely(kvm_is_error_hva(hva)))
189 return false;
190
191 /*
192 * do alloc nowait since if we are going to sleep anyway we
193 * may as well sleep faulting in page
194 */
195 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
196 if (!work)
197 return false;
198
199 work->wakeup_all = false;
200 work->vcpu = vcpu;
201 work->cr2_or_gpa = cr2_or_gpa;
202 work->addr = hva;
203 work->arch = *arch;
204
205 INIT_WORK(&work->work, async_pf_execute);
206
207 list_add_tail(&work->queue, &vcpu->async_pf.queue);
208 vcpu->async_pf.queued++;
209 work->notpresent_injected = kvm_arch_async_page_not_present(vcpu, work);
210
211 schedule_work(&work->work);
212
213 return true;
214}
215
216int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
217{
218 struct kvm_async_pf *work;
219 bool first;
220
221 if (!list_empty_careful(&vcpu->async_pf.done))
222 return 0;
223
224 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
225 if (!work)
226 return -ENOMEM;
227
228 work->wakeup_all = true;
229 INIT_LIST_HEAD(&work->queue); /* for list_del to work */
230
231 spin_lock(&vcpu->async_pf.lock);
232 first = list_empty(&vcpu->async_pf.done);
233 list_add_tail(&work->link, &vcpu->async_pf.done);
234 spin_unlock(&vcpu->async_pf.lock);
235
236 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
237 kvm_arch_async_page_present_queued(vcpu);
238
239 vcpu->async_pf.queued++;
240 return 0;
241}
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 struct kmem_cache *async_pf_cache;
32
33int 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
43void 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
50void 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
57static 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
95void 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
123void 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
147int 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;
188retry_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
195int 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}