1af585b92SGleb Natapov /* 2af585b92SGleb Natapov * kvm asynchronous fault support 3af585b92SGleb Natapov * 4af585b92SGleb Natapov * Copyright 2010 Red Hat, Inc. 5af585b92SGleb Natapov * 6af585b92SGleb Natapov * Author: 7af585b92SGleb Natapov * Gleb Natapov <gleb@redhat.com> 8af585b92SGleb Natapov * 9af585b92SGleb Natapov * This file is free software; you can redistribute it and/or modify 10af585b92SGleb Natapov * it under the terms of version 2 of the GNU General Public License 11af585b92SGleb Natapov * as published by the Free Software Foundation. 12af585b92SGleb Natapov * 13af585b92SGleb Natapov * This program is distributed in the hope that it will be useful, 14af585b92SGleb Natapov * but WITHOUT ANY WARRANTY; without even the implied warranty of 15af585b92SGleb Natapov * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16af585b92SGleb Natapov * GNU General Public License for more details. 17af585b92SGleb Natapov * 18af585b92SGleb Natapov * You should have received a copy of the GNU General Public License 19af585b92SGleb Natapov * along with this program; if not, write to the Free Software Foundation, 20af585b92SGleb Natapov * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. 21af585b92SGleb Natapov */ 22af585b92SGleb Natapov 23af585b92SGleb Natapov #include <linux/kvm_host.h> 24af585b92SGleb Natapov #include <linux/slab.h> 25af585b92SGleb Natapov #include <linux/module.h> 26af585b92SGleb Natapov #include <linux/mmu_context.h> 27af585b92SGleb Natapov 28af585b92SGleb Natapov #include "async_pf.h" 29af585b92SGleb Natapov #include <trace/events/kvm.h> 30af585b92SGleb Natapov 31e0ead41aSDominik Dingel static inline void kvm_async_page_present_sync(struct kvm_vcpu *vcpu, 32e0ead41aSDominik Dingel struct kvm_async_pf *work) 33e0ead41aSDominik Dingel { 34e0ead41aSDominik Dingel #ifdef CONFIG_KVM_ASYNC_PF_SYNC 35e0ead41aSDominik Dingel kvm_arch_async_page_present(vcpu, work); 36e0ead41aSDominik Dingel #endif 37e0ead41aSDominik Dingel } 38e0ead41aSDominik Dingel static inline void kvm_async_page_present_async(struct kvm_vcpu *vcpu, 39e0ead41aSDominik Dingel struct kvm_async_pf *work) 40e0ead41aSDominik Dingel { 41e0ead41aSDominik Dingel #ifndef CONFIG_KVM_ASYNC_PF_SYNC 42e0ead41aSDominik Dingel kvm_arch_async_page_present(vcpu, work); 43e0ead41aSDominik Dingel #endif 44e0ead41aSDominik Dingel } 45e0ead41aSDominik Dingel 46af585b92SGleb Natapov static struct kmem_cache *async_pf_cache; 47af585b92SGleb Natapov 48af585b92SGleb Natapov int kvm_async_pf_init(void) 49af585b92SGleb Natapov { 50af585b92SGleb Natapov async_pf_cache = KMEM_CACHE(kvm_async_pf, 0); 51af585b92SGleb Natapov 52af585b92SGleb Natapov if (!async_pf_cache) 53af585b92SGleb Natapov return -ENOMEM; 54af585b92SGleb Natapov 55af585b92SGleb Natapov return 0; 56af585b92SGleb Natapov } 57af585b92SGleb Natapov 58af585b92SGleb Natapov void kvm_async_pf_deinit(void) 59af585b92SGleb Natapov { 60af585b92SGleb Natapov kmem_cache_destroy(async_pf_cache); 61af585b92SGleb Natapov async_pf_cache = NULL; 62af585b92SGleb Natapov } 63af585b92SGleb Natapov 64af585b92SGleb Natapov void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu) 65af585b92SGleb Natapov { 66af585b92SGleb Natapov INIT_LIST_HEAD(&vcpu->async_pf.done); 67af585b92SGleb Natapov INIT_LIST_HEAD(&vcpu->async_pf.queue); 68af585b92SGleb Natapov spin_lock_init(&vcpu->async_pf.lock); 69af585b92SGleb Natapov } 70af585b92SGleb Natapov 71af585b92SGleb Natapov static void async_pf_execute(struct work_struct *work) 72af585b92SGleb Natapov { 73af585b92SGleb Natapov struct kvm_async_pf *apf = 74af585b92SGleb Natapov container_of(work, struct kvm_async_pf, work); 75af585b92SGleb Natapov struct mm_struct *mm = apf->mm; 76af585b92SGleb Natapov struct kvm_vcpu *vcpu = apf->vcpu; 77af585b92SGleb Natapov unsigned long addr = apf->addr; 78af585b92SGleb Natapov gva_t gva = apf->gva; 79af585b92SGleb Natapov 80af585b92SGleb Natapov might_sleep(); 81af585b92SGleb Natapov 82*1e987790SDave Hansen /* 83*1e987790SDave Hansen * This work is run asynchromously to the task which owns 84*1e987790SDave Hansen * mm and might be done in another context, so we must 85*1e987790SDave Hansen * use FOLL_REMOTE. 86*1e987790SDave Hansen */ 87*1e987790SDave Hansen __get_user_pages_unlocked(NULL, mm, addr, 1, 1, 0, NULL, FOLL_REMOTE); 88*1e987790SDave Hansen 89e0ead41aSDominik Dingel kvm_async_page_present_sync(vcpu, apf); 90af585b92SGleb Natapov 91af585b92SGleb Natapov spin_lock(&vcpu->async_pf.lock); 92af585b92SGleb Natapov list_add_tail(&apf->link, &vcpu->async_pf.done); 93af585b92SGleb Natapov spin_unlock(&vcpu->async_pf.lock); 94af585b92SGleb Natapov 95af585b92SGleb Natapov /* 96af585b92SGleb Natapov * apf may be freed by kvm_check_async_pf_completion() after 97af585b92SGleb Natapov * this point 98af585b92SGleb Natapov */ 99af585b92SGleb Natapov 100f2e10669Schai wen trace_kvm_async_pf_completed(addr, gva); 101af585b92SGleb Natapov 1026003a420SKosuke Tatsukawa /* 1036003a420SKosuke Tatsukawa * This memory barrier pairs with prepare_to_wait's set_current_state() 1046003a420SKosuke Tatsukawa */ 1056003a420SKosuke Tatsukawa smp_mb(); 106af585b92SGleb Natapov if (waitqueue_active(&vcpu->wq)) 107af585b92SGleb Natapov wake_up_interruptible(&vcpu->wq); 108af585b92SGleb Natapov 10941c22f62SOleg Nesterov mmput(mm); 110af585b92SGleb Natapov kvm_put_kvm(vcpu->kvm); 111af585b92SGleb Natapov } 112af585b92SGleb Natapov 113af585b92SGleb Natapov void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) 114af585b92SGleb Natapov { 115af585b92SGleb Natapov /* cancel outstanding work queue item */ 116af585b92SGleb Natapov while (!list_empty(&vcpu->async_pf.queue)) { 117af585b92SGleb Natapov struct kvm_async_pf *work = 118af585b92SGleb Natapov list_entry(vcpu->async_pf.queue.next, 119af585b92SGleb Natapov typeof(*work), queue); 120af585b92SGleb Natapov list_del(&work->queue); 1219f2ceda4SDominik Dingel 1229f2ceda4SDominik Dingel #ifdef CONFIG_KVM_ASYNC_PF_SYNC 1239f2ceda4SDominik Dingel flush_work(&work->work); 1249f2ceda4SDominik Dingel #else 12598fda169SRadim Krčmář if (cancel_work_sync(&work->work)) { 12641c22f62SOleg Nesterov mmput(work->mm); 12728b441e2SRadim Krčmář kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */ 128af585b92SGleb Natapov kmem_cache_free(async_pf_cache, work); 129af585b92SGleb Natapov } 1309f2ceda4SDominik Dingel #endif 13128b441e2SRadim Krčmář } 132af585b92SGleb Natapov 133af585b92SGleb Natapov spin_lock(&vcpu->async_pf.lock); 134af585b92SGleb Natapov while (!list_empty(&vcpu->async_pf.done)) { 135af585b92SGleb Natapov struct kvm_async_pf *work = 136af585b92SGleb Natapov list_entry(vcpu->async_pf.done.next, 137af585b92SGleb Natapov typeof(*work), link); 138af585b92SGleb Natapov list_del(&work->link); 139af585b92SGleb Natapov kmem_cache_free(async_pf_cache, work); 140af585b92SGleb Natapov } 141af585b92SGleb Natapov spin_unlock(&vcpu->async_pf.lock); 142af585b92SGleb Natapov 143af585b92SGleb Natapov vcpu->async_pf.queued = 0; 144af585b92SGleb Natapov } 145af585b92SGleb Natapov 146af585b92SGleb Natapov void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu) 147af585b92SGleb Natapov { 148af585b92SGleb Natapov struct kvm_async_pf *work; 149af585b92SGleb Natapov 15015096ffcSXiao Guangrong while (!list_empty_careful(&vcpu->async_pf.done) && 15115096ffcSXiao Guangrong kvm_arch_can_inject_async_page_present(vcpu)) { 152af585b92SGleb Natapov spin_lock(&vcpu->async_pf.lock); 15315096ffcSXiao Guangrong work = list_first_entry(&vcpu->async_pf.done, typeof(*work), 15415096ffcSXiao Guangrong link); 155af585b92SGleb Natapov list_del(&work->link); 156af585b92SGleb Natapov spin_unlock(&vcpu->async_pf.lock); 157af585b92SGleb Natapov 15856028d08SGleb Natapov kvm_arch_async_page_ready(vcpu, work); 1591179ba53SDominik Dingel kvm_async_page_present_async(vcpu, work); 160af585b92SGleb Natapov 161af585b92SGleb Natapov list_del(&work->queue); 162af585b92SGleb Natapov vcpu->async_pf.queued--; 163af585b92SGleb Natapov kmem_cache_free(async_pf_cache, work); 164af585b92SGleb Natapov } 16515096ffcSXiao Guangrong } 166af585b92SGleb Natapov 167e0ead41aSDominik Dingel int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva, 168af585b92SGleb Natapov struct kvm_arch_async_pf *arch) 169af585b92SGleb Natapov { 170af585b92SGleb Natapov struct kvm_async_pf *work; 171af585b92SGleb Natapov 172af585b92SGleb Natapov if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU) 173af585b92SGleb Natapov return 0; 174af585b92SGleb Natapov 175af585b92SGleb Natapov /* setup delayed work */ 176af585b92SGleb Natapov 177af585b92SGleb Natapov /* 178af585b92SGleb Natapov * do alloc nowait since if we are going to sleep anyway we 179af585b92SGleb Natapov * may as well sleep faulting in page 180af585b92SGleb Natapov */ 181af585b92SGleb Natapov work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT); 182af585b92SGleb Natapov if (!work) 183af585b92SGleb Natapov return 0; 184af585b92SGleb Natapov 185f2e10669Schai wen work->wakeup_all = false; 186af585b92SGleb Natapov work->vcpu = vcpu; 187af585b92SGleb Natapov work->gva = gva; 188e0ead41aSDominik Dingel work->addr = hva; 189af585b92SGleb Natapov work->arch = *arch; 190af585b92SGleb Natapov work->mm = current->mm; 19141c22f62SOleg Nesterov atomic_inc(&work->mm->mm_users); 192af585b92SGleb Natapov kvm_get_kvm(work->vcpu->kvm); 193af585b92SGleb Natapov 194af585b92SGleb Natapov /* this can't really happen otherwise gfn_to_pfn_async 195af585b92SGleb Natapov would succeed */ 196af585b92SGleb Natapov if (unlikely(kvm_is_error_hva(work->addr))) 197af585b92SGleb Natapov goto retry_sync; 198af585b92SGleb Natapov 199af585b92SGleb Natapov INIT_WORK(&work->work, async_pf_execute); 200af585b92SGleb Natapov if (!schedule_work(&work->work)) 201af585b92SGleb Natapov goto retry_sync; 202af585b92SGleb Natapov 203af585b92SGleb Natapov list_add_tail(&work->queue, &vcpu->async_pf.queue); 204af585b92SGleb Natapov vcpu->async_pf.queued++; 205af585b92SGleb Natapov kvm_arch_async_page_not_present(vcpu, work); 206af585b92SGleb Natapov return 1; 207af585b92SGleb Natapov retry_sync: 208af585b92SGleb Natapov kvm_put_kvm(work->vcpu->kvm); 20941c22f62SOleg Nesterov mmput(work->mm); 210af585b92SGleb Natapov kmem_cache_free(async_pf_cache, work); 211af585b92SGleb Natapov return 0; 212af585b92SGleb Natapov } 213344d9588SGleb Natapov 214344d9588SGleb Natapov int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu) 215344d9588SGleb Natapov { 216344d9588SGleb Natapov struct kvm_async_pf *work; 217344d9588SGleb Natapov 21864f638c7SXiao Guangrong if (!list_empty_careful(&vcpu->async_pf.done)) 219344d9588SGleb Natapov return 0; 220344d9588SGleb Natapov 221344d9588SGleb Natapov work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC); 222344d9588SGleb Natapov if (!work) 223344d9588SGleb Natapov return -ENOMEM; 224344d9588SGleb Natapov 225f2e10669Schai wen work->wakeup_all = true; 226344d9588SGleb Natapov INIT_LIST_HEAD(&work->queue); /* for list_del to work */ 227344d9588SGleb Natapov 22864f638c7SXiao Guangrong spin_lock(&vcpu->async_pf.lock); 229344d9588SGleb Natapov list_add_tail(&work->link, &vcpu->async_pf.done); 23064f638c7SXiao Guangrong spin_unlock(&vcpu->async_pf.lock); 23164f638c7SXiao Guangrong 232344d9588SGleb Natapov vcpu->async_pf.queued++; 233344d9588SGleb Natapov return 0; 234344d9588SGleb Natapov } 235