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
20 static struct kmem_cache *async_pf_cache;
21
kvm_async_pf_init(void)22 int 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
kvm_async_pf_deinit(void)32 void kvm_async_pf_deinit(void)
33 {
34 kmem_cache_destroy(async_pf_cache);
35 async_pf_cache = NULL;
36 }
37
kvm_async_pf_vcpu_init(struct kvm_vcpu * vcpu)38 void 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
async_pf_execute(struct work_struct * work)45 static 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 mm_struct *mm = apf->mm;
50 struct kvm_vcpu *vcpu = apf->vcpu;
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 * This work is run asynchronously to the task which owns
60 * mm and might be done in another context, so we must
61 * access remotely.
62 */
63 mmap_read_lock(mm);
64 get_user_pages_remote(mm, addr, 1, FOLL_WRITE, NULL, &locked);
65 if (locked)
66 mmap_read_unlock(mm);
67
68 if (IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
69 kvm_arch_async_page_present(vcpu, apf);
70
71 spin_lock(&vcpu->async_pf.lock);
72 first = list_empty(&vcpu->async_pf.done);
73 list_add_tail(&apf->link, &vcpu->async_pf.done);
74 apf->vcpu = NULL;
75 spin_unlock(&vcpu->async_pf.lock);
76
77 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
78 kvm_arch_async_page_present_queued(vcpu);
79
80 /*
81 * apf may be freed by kvm_check_async_pf_completion() after
82 * this point
83 */
84
85 trace_kvm_async_pf_completed(addr, cr2_or_gpa);
86
87 __kvm_vcpu_wake_up(vcpu);
88
89 mmput(mm);
90 }
91
kvm_flush_and_free_async_pf_work(struct kvm_async_pf * work)92 static void kvm_flush_and_free_async_pf_work(struct kvm_async_pf *work)
93 {
94 /*
95 * The async #PF is "done", but KVM must wait for the work item itself,
96 * i.e. async_pf_execute(), to run to completion. If KVM is a module,
97 * KVM must ensure *no* code owned by the KVM (the module) can be run
98 * after the last call to module_put(). Note, flushing the work item
99 * is always required when the item is taken off the completion queue.
100 * E.g. even if the vCPU handles the item in the "normal" path, the VM
101 * could be terminated before async_pf_execute() completes.
102 *
103 * Wake all events skip the queue and go straight done, i.e. don't
104 * need to be flushed (but sanity check that the work wasn't queued).
105 */
106 if (work->wakeup_all)
107 WARN_ON_ONCE(work->work.func);
108 else
109 flush_work(&work->work);
110 kmem_cache_free(async_pf_cache, work);
111 }
112
kvm_clear_async_pf_completion_queue(struct kvm_vcpu * vcpu)113 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
114 {
115 spin_lock(&vcpu->async_pf.lock);
116
117 /* cancel outstanding work queue item */
118 while (!list_empty(&vcpu->async_pf.queue)) {
119 struct kvm_async_pf *work =
120 list_first_entry(&vcpu->async_pf.queue,
121 typeof(*work), queue);
122 list_del(&work->queue);
123
124 /*
125 * We know it's present in vcpu->async_pf.done, do
126 * nothing here.
127 */
128 if (!work->vcpu)
129 continue;
130
131 spin_unlock(&vcpu->async_pf.lock);
132 #ifdef CONFIG_KVM_ASYNC_PF_SYNC
133 flush_work(&work->work);
134 #else
135 if (cancel_work_sync(&work->work)) {
136 mmput(work->mm);
137 kmem_cache_free(async_pf_cache, work);
138 }
139 #endif
140 spin_lock(&vcpu->async_pf.lock);
141 }
142
143 while (!list_empty(&vcpu->async_pf.done)) {
144 struct kvm_async_pf *work =
145 list_first_entry(&vcpu->async_pf.done,
146 typeof(*work), link);
147 list_del(&work->link);
148
149 spin_unlock(&vcpu->async_pf.lock);
150 kvm_flush_and_free_async_pf_work(work);
151 spin_lock(&vcpu->async_pf.lock);
152 }
153 spin_unlock(&vcpu->async_pf.lock);
154
155 vcpu->async_pf.queued = 0;
156 }
157
kvm_check_async_pf_completion(struct kvm_vcpu * vcpu)158 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
159 {
160 struct kvm_async_pf *work;
161
162 while (!list_empty_careful(&vcpu->async_pf.done) &&
163 kvm_arch_can_dequeue_async_page_present(vcpu)) {
164 spin_lock(&vcpu->async_pf.lock);
165 work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
166 link);
167 list_del(&work->link);
168 spin_unlock(&vcpu->async_pf.lock);
169
170 kvm_arch_async_page_ready(vcpu, work);
171 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
172 kvm_arch_async_page_present(vcpu, work);
173
174 list_del(&work->queue);
175 vcpu->async_pf.queued--;
176 kvm_flush_and_free_async_pf_work(work);
177 }
178 }
179
180 /*
181 * Try to schedule a job to handle page fault asynchronously. Returns 'true' on
182 * success, 'false' on failure (page fault has to be handled synchronously).
183 */
kvm_setup_async_pf(struct kvm_vcpu * vcpu,gpa_t cr2_or_gpa,unsigned long hva,struct kvm_arch_async_pf * arch)184 bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
185 unsigned long hva, struct kvm_arch_async_pf *arch)
186 {
187 struct kvm_async_pf *work;
188
189 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
190 return false;
191
192 /* Arch specific code should not do async PF in this case */
193 if (unlikely(kvm_is_error_hva(hva)))
194 return false;
195
196 /*
197 * do alloc nowait since if we are going to sleep anyway we
198 * may as well sleep faulting in page
199 */
200 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
201 if (!work)
202 return false;
203
204 work->wakeup_all = false;
205 work->vcpu = vcpu;
206 work->cr2_or_gpa = cr2_or_gpa;
207 work->addr = hva;
208 work->arch = *arch;
209 work->mm = current->mm;
210 mmget(work->mm);
211
212 INIT_WORK(&work->work, async_pf_execute);
213
214 list_add_tail(&work->queue, &vcpu->async_pf.queue);
215 vcpu->async_pf.queued++;
216 work->notpresent_injected = kvm_arch_async_page_not_present(vcpu, work);
217
218 schedule_work(&work->work);
219
220 return true;
221 }
222
kvm_async_pf_wakeup_all(struct kvm_vcpu * vcpu)223 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
224 {
225 struct kvm_async_pf *work;
226 bool first;
227
228 if (!list_empty_careful(&vcpu->async_pf.done))
229 return 0;
230
231 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
232 if (!work)
233 return -ENOMEM;
234
235 work->wakeup_all = true;
236 INIT_LIST_HEAD(&work->queue); /* for list_del to work */
237
238 spin_lock(&vcpu->async_pf.lock);
239 first = list_empty(&vcpu->async_pf.done);
240 list_add_tail(&work->link, &vcpu->async_pf.done);
241 spin_unlock(&vcpu->async_pf.lock);
242
243 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
244 kvm_arch_async_page_present_queued(vcpu);
245
246 vcpu->async_pf.queued++;
247 return 0;
248 }
249