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
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
32 void kvm_async_pf_deinit(void)
33 {
34 kmem_cache_destroy(async_pf_cache);
35 async_pf_cache = NULL;
36 }
37
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
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 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
99 static 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
120 void 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
153 void 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 */
179 bool 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
216 int 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 }
242
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.