| 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 | |
| 31 | static 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 | } |
| 38 | static 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 | |
| 46 | static struct kmem_cache *async_pf_cache; |
| 47 | |
| 48 | int 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 | |
| 58 | void 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 | |
| 65 | void 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 | |
| 72 | static 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 | /* |
| 98 | * This memory barrier pairs with prepare_to_wait's set_current_state() |
| 99 | */ |
| 100 | smp_mb(); |
| 101 | if (waitqueue_active(&vcpu->wq)) |
| 102 | wake_up_interruptible(&vcpu->wq); |
| 103 | |
| 104 | mmput(mm); |
| 105 | kvm_put_kvm(vcpu->kvm); |
| 106 | } |
| 107 | |
| 108 | void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) |
| 109 | { |
| 110 | /* cancel outstanding work queue item */ |
| 111 | while (!list_empty(&vcpu->async_pf.queue)) { |
| 112 | struct kvm_async_pf *work = |
| 113 | list_entry(vcpu->async_pf.queue.next, |
| 114 | typeof(*work), queue); |
| 115 | list_del(&work->queue); |
| 116 | |
| 117 | #ifdef CONFIG_KVM_ASYNC_PF_SYNC |
| 118 | flush_work(&work->work); |
| 119 | #else |
| 120 | if (cancel_work_sync(&work->work)) { |
| 121 | mmput(work->mm); |
| 122 | kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */ |
| 123 | kmem_cache_free(async_pf_cache, work); |
| 124 | } |
| 125 | #endif |
| 126 | } |
| 127 | |
| 128 | spin_lock(&vcpu->async_pf.lock); |
| 129 | while (!list_empty(&vcpu->async_pf.done)) { |
| 130 | struct kvm_async_pf *work = |
| 131 | list_entry(vcpu->async_pf.done.next, |
| 132 | typeof(*work), link); |
| 133 | list_del(&work->link); |
| 134 | kmem_cache_free(async_pf_cache, work); |
| 135 | } |
| 136 | spin_unlock(&vcpu->async_pf.lock); |
| 137 | |
| 138 | vcpu->async_pf.queued = 0; |
| 139 | } |
| 140 | |
| 141 | void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu) |
| 142 | { |
| 143 | struct kvm_async_pf *work; |
| 144 | |
| 145 | while (!list_empty_careful(&vcpu->async_pf.done) && |
| 146 | kvm_arch_can_inject_async_page_present(vcpu)) { |
| 147 | spin_lock(&vcpu->async_pf.lock); |
| 148 | work = list_first_entry(&vcpu->async_pf.done, typeof(*work), |
| 149 | link); |
| 150 | list_del(&work->link); |
| 151 | spin_unlock(&vcpu->async_pf.lock); |
| 152 | |
| 153 | kvm_arch_async_page_ready(vcpu, work); |
| 154 | kvm_async_page_present_async(vcpu, work); |
| 155 | |
| 156 | list_del(&work->queue); |
| 157 | vcpu->async_pf.queued--; |
| 158 | kmem_cache_free(async_pf_cache, work); |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva, |
| 163 | struct kvm_arch_async_pf *arch) |
| 164 | { |
| 165 | struct kvm_async_pf *work; |
| 166 | |
| 167 | if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU) |
| 168 | return 0; |
| 169 | |
| 170 | /* setup delayed work */ |
| 171 | |
| 172 | /* |
| 173 | * do alloc nowait since if we are going to sleep anyway we |
| 174 | * may as well sleep faulting in page |
| 175 | */ |
| 176 | work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN); |
| 177 | if (!work) |
| 178 | return 0; |
| 179 | |
| 180 | work->wakeup_all = false; |
| 181 | work->vcpu = vcpu; |
| 182 | work->gva = gva; |
| 183 | work->addr = hva; |
| 184 | work->arch = *arch; |
| 185 | work->mm = current->mm; |
| 186 | atomic_inc(&work->mm->mm_users); |
| 187 | kvm_get_kvm(work->vcpu->kvm); |
| 188 | |
| 189 | /* this can't really happen otherwise gfn_to_pfn_async |
| 190 | would succeed */ |
| 191 | if (unlikely(kvm_is_error_hva(work->addr))) |
| 192 | goto retry_sync; |
| 193 | |
| 194 | INIT_WORK(&work->work, async_pf_execute); |
| 195 | if (!schedule_work(&work->work)) |
| 196 | goto retry_sync; |
| 197 | |
| 198 | list_add_tail(&work->queue, &vcpu->async_pf.queue); |
| 199 | vcpu->async_pf.queued++; |
| 200 | kvm_arch_async_page_not_present(vcpu, work); |
| 201 | return 1; |
| 202 | retry_sync: |
| 203 | kvm_put_kvm(work->vcpu->kvm); |
| 204 | mmput(work->mm); |
| 205 | kmem_cache_free(async_pf_cache, work); |
| 206 | return 0; |
| 207 | } |
| 208 | |
| 209 | int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu) |
| 210 | { |
| 211 | struct kvm_async_pf *work; |
| 212 | |
| 213 | if (!list_empty_careful(&vcpu->async_pf.done)) |
| 214 | return 0; |
| 215 | |
| 216 | work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC); |
| 217 | if (!work) |
| 218 | return -ENOMEM; |
| 219 | |
| 220 | work->wakeup_all = true; |
| 221 | INIT_LIST_HEAD(&work->queue); /* for list_del to work */ |
| 222 | |
| 223 | spin_lock(&vcpu->async_pf.lock); |
| 224 | list_add_tail(&work->link, &vcpu->async_pf.done); |
| 225 | spin_unlock(&vcpu->async_pf.lock); |
| 226 | |
| 227 | vcpu->async_pf.queued++; |
| 228 | return 0; |
| 229 | } |