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 kmem_cache_destroy(async_pf_cache); 61 async_pf_cache = NULL; 62 } 63 64 void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu) 65 { 66 INIT_LIST_HEAD(&vcpu->async_pf.done); 67 INIT_LIST_HEAD(&vcpu->async_pf.queue); 68 spin_lock_init(&vcpu->async_pf.lock); 69 } 70 71 static void async_pf_execute(struct work_struct *work) 72 { 73 struct kvm_async_pf *apf = 74 container_of(work, struct kvm_async_pf, work); 75 struct mm_struct *mm = apf->mm; 76 struct kvm_vcpu *vcpu = apf->vcpu; 77 unsigned long addr = apf->addr; 78 gva_t gva = apf->gva; 79 80 might_sleep(); 81 82 /* 83 * This work is run asynchromously to the task which owns 84 * mm and might be done in another context, so we must 85 * use FOLL_REMOTE. 86 */ 87 __get_user_pages_unlocked(NULL, mm, addr, 1, 1, 0, NULL, FOLL_REMOTE); 88 89 kvm_async_page_present_sync(vcpu, apf); 90 91 spin_lock(&vcpu->async_pf.lock); 92 list_add_tail(&apf->link, &vcpu->async_pf.done); 93 spin_unlock(&vcpu->async_pf.lock); 94 95 /* 96 * apf may be freed by kvm_check_async_pf_completion() after 97 * this point 98 */ 99 100 trace_kvm_async_pf_completed(addr, gva); 101 102 /* 103 * This memory barrier pairs with prepare_to_wait's set_current_state() 104 */ 105 smp_mb(); 106 if (swait_active(&vcpu->wq)) 107 swake_up(&vcpu->wq); 108 109 mmput(mm); 110 kvm_put_kvm(vcpu->kvm); 111 } 112 113 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) 114 { 115 /* cancel outstanding work queue item */ 116 while (!list_empty(&vcpu->async_pf.queue)) { 117 struct kvm_async_pf *work = 118 list_first_entry(&vcpu->async_pf.queue, 119 typeof(*work), queue); 120 list_del(&work->queue); 121 122 #ifdef CONFIG_KVM_ASYNC_PF_SYNC 123 flush_work(&work->work); 124 #else 125 if (cancel_work_sync(&work->work)) { 126 mmput(work->mm); 127 kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */ 128 kmem_cache_free(async_pf_cache, work); 129 } 130 #endif 131 } 132 133 spin_lock(&vcpu->async_pf.lock); 134 while (!list_empty(&vcpu->async_pf.done)) { 135 struct kvm_async_pf *work = 136 list_first_entry(&vcpu->async_pf.done, 137 typeof(*work), link); 138 list_del(&work->link); 139 kmem_cache_free(async_pf_cache, work); 140 } 141 spin_unlock(&vcpu->async_pf.lock); 142 143 vcpu->async_pf.queued = 0; 144 } 145 146 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu) 147 { 148 struct kvm_async_pf *work; 149 150 while (!list_empty_careful(&vcpu->async_pf.done) && 151 kvm_arch_can_inject_async_page_present(vcpu)) { 152 spin_lock(&vcpu->async_pf.lock); 153 work = list_first_entry(&vcpu->async_pf.done, typeof(*work), 154 link); 155 list_del(&work->link); 156 spin_unlock(&vcpu->async_pf.lock); 157 158 kvm_arch_async_page_ready(vcpu, work); 159 kvm_async_page_present_async(vcpu, work); 160 161 list_del(&work->queue); 162 vcpu->async_pf.queued--; 163 kmem_cache_free(async_pf_cache, work); 164 } 165 } 166 167 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva, 168 struct kvm_arch_async_pf *arch) 169 { 170 struct kvm_async_pf *work; 171 172 if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU) 173 return 0; 174 175 /* setup delayed work */ 176 177 /* 178 * do alloc nowait since if we are going to sleep anyway we 179 * may as well sleep faulting in page 180 */ 181 work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN); 182 if (!work) 183 return 0; 184 185 work->wakeup_all = false; 186 work->vcpu = vcpu; 187 work->gva = gva; 188 work->addr = hva; 189 work->arch = *arch; 190 work->mm = current->mm; 191 atomic_inc(&work->mm->mm_users); 192 kvm_get_kvm(work->vcpu->kvm); 193 194 /* this can't really happen otherwise gfn_to_pfn_async 195 would succeed */ 196 if (unlikely(kvm_is_error_hva(work->addr))) 197 goto retry_sync; 198 199 INIT_WORK(&work->work, async_pf_execute); 200 if (!schedule_work(&work->work)) 201 goto retry_sync; 202 203 list_add_tail(&work->queue, &vcpu->async_pf.queue); 204 vcpu->async_pf.queued++; 205 kvm_arch_async_page_not_present(vcpu, work); 206 return 1; 207 retry_sync: 208 kvm_put_kvm(work->vcpu->kvm); 209 mmput(work->mm); 210 kmem_cache_free(async_pf_cache, work); 211 return 0; 212 } 213 214 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu) 215 { 216 struct kvm_async_pf *work; 217 218 if (!list_empty_careful(&vcpu->async_pf.done)) 219 return 0; 220 221 work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC); 222 if (!work) 223 return -ENOMEM; 224 225 work->wakeup_all = true; 226 INIT_LIST_HEAD(&work->queue); /* for list_del to work */ 227 228 spin_lock(&vcpu->async_pf.lock); 229 list_add_tail(&work->link, &vcpu->async_pf.done); 230 spin_unlock(&vcpu->async_pf.lock); 231 232 vcpu->async_pf.queued++; 233 return 0; 234 } 235