1 // SPDX-License-Identifier: GPL-2.0-only
3 * kvm asynchronous fault support
5 * Copyright 2010 Red Hat, Inc.
8 * Gleb Natapov <gleb@redhat.com>
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>
18 #include <trace/events/kvm.h>
20 static struct kmem_cache *async_pf_cache;
22 int kvm_async_pf_init(void)
24 async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
32 void kvm_async_pf_deinit(void)
34 kmem_cache_destroy(async_pf_cache);
35 async_pf_cache = NULL;
38 void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
40 INIT_LIST_HEAD(&vcpu->async_pf.done);
41 INIT_LIST_HEAD(&vcpu->async_pf.queue);
42 spin_lock_init(&vcpu->async_pf.lock);
45 static void async_pf_execute(struct work_struct *work)
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;
59 * This work is run asynchronously to the task which owns
60 * mm and might be done in another context, so we must
64 get_user_pages_remote(mm, addr, 1, FOLL_WRITE, NULL, &locked);
68 if (IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
69 kvm_arch_async_page_present(vcpu, apf);
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);
75 spin_unlock(&vcpu->async_pf.lock);
77 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
78 kvm_arch_async_page_present_queued(vcpu);
81 * apf may be freed by kvm_check_async_pf_completion() after
85 trace_kvm_async_pf_completed(addr, cr2_or_gpa);
87 __kvm_vcpu_wake_up(vcpu);
92 static void kvm_flush_and_free_async_pf_work(struct kvm_async_pf *work)
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.
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).
106 if (work->wakeup_all)
107 WARN_ON_ONCE(work->work.func);
109 flush_work(&work->work);
110 kmem_cache_free(async_pf_cache, work);
113 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
115 spin_lock(&vcpu->async_pf.lock);
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);
125 * We know it's present in vcpu->async_pf.done, do
131 spin_unlock(&vcpu->async_pf.lock);
132 #ifdef CONFIG_KVM_ASYNC_PF_SYNC
133 flush_work(&work->work);
135 if (cancel_work_sync(&work->work)) {
137 kmem_cache_free(async_pf_cache, work);
140 spin_lock(&vcpu->async_pf.lock);
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);
149 spin_unlock(&vcpu->async_pf.lock);
150 kvm_flush_and_free_async_pf_work(work);
151 spin_lock(&vcpu->async_pf.lock);
153 spin_unlock(&vcpu->async_pf.lock);
155 vcpu->async_pf.queued = 0;
158 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
160 struct kvm_async_pf *work;
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),
167 list_del(&work->link);
168 spin_unlock(&vcpu->async_pf.lock);
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);
174 list_del(&work->queue);
175 vcpu->async_pf.queued--;
176 kvm_flush_and_free_async_pf_work(work);
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).
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)
187 struct kvm_async_pf *work;
189 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
192 /* Arch specific code should not do async PF in this case */
193 if (unlikely(kvm_is_error_hva(hva)))
197 * do alloc nowait since if we are going to sleep anyway we
198 * may as well sleep faulting in page
200 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
204 work->wakeup_all = false;
206 work->cr2_or_gpa = cr2_or_gpa;
209 work->mm = current->mm;
212 INIT_WORK(&work->work, async_pf_execute);
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);
218 schedule_work(&work->work);
223 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
225 struct kvm_async_pf *work;
228 if (!list_empty_careful(&vcpu->async_pf.done))
231 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
235 work->wakeup_all = true;
236 INIT_LIST_HEAD(&work->queue); /* for list_del to work */
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);
243 if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
244 kvm_arch_async_page_present_queued(vcpu);
246 vcpu->async_pf.queued++;