From: Paolo Bonzini Date: Thu, 17 Nov 2016 14:55:45 +0000 (+0100) Subject: KVM: async_pf: avoid recursive flushing of work items X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=22583f0d9c85e60c9860bc8a0ebff59fe08be6d7;p=GitHub%2Fmoto-9609%2Fandroid_kernel_motorola_exynos9610.git KVM: async_pf: avoid recursive flushing of work items This was reported by syzkaller: [ INFO: possible recursive locking detected ] 4.9.0-rc4+ #49 Not tainted --------------------------------------------- kworker/2:1/5658 is trying to acquire lock: ([ 1644.769018] (&work->work) [< inline >] list_empty include/linux/compiler.h:243 [] flush_work+0x0/0x660 kernel/workqueue.c:1511 but task is already holding lock: ([ 1644.769018] (&work->work) [] process_one_work+0x94b/0x1900 kernel/workqueue.c:2093 stack backtrace: CPU: 2 PID: 5658 Comm: kworker/2:1 Not tainted 4.9.0-rc4+ #49 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011 Workqueue: events async_pf_execute ffff8800676ff630 ffffffff81c2e46b ffffffff8485b930 ffff88006b1fc480 0000000000000000 ffffffff8485b930 ffff8800676ff7e0 ffffffff81339b27 ffff8800676ff7e8 0000000000000046 ffff88006b1fcce8 ffff88006b1fccf0 Call Trace: ... [] flush_work+0x93/0x660 kernel/workqueue.c:2846 [] __cancel_work_timer+0x17a/0x410 kernel/workqueue.c:2916 [] cancel_work_sync+0x17/0x20 kernel/workqueue.c:2951 [] kvm_clear_async_pf_completion_queue+0xd7/0x400 virt/kvm/async_pf.c:126 [< inline >] kvm_free_vcpus arch/x86/kvm/x86.c:7841 [] kvm_arch_destroy_vm+0x23d/0x620 arch/x86/kvm/x86.c:7946 [< inline >] kvm_destroy_vm virt/kvm/kvm_main.c:731 [] kvm_put_kvm+0x40e/0x790 virt/kvm/kvm_main.c:752 [] async_pf_execute+0x23d/0x4f0 virt/kvm/async_pf.c:111 [] process_one_work+0x9fc/0x1900 kernel/workqueue.c:2096 [] worker_thread+0xef/0x1480 kernel/workqueue.c:2230 [] kthread+0x244/0x2d0 kernel/kthread.c:209 [] ret_from_fork+0x2a/0x40 arch/x86/entry/entry_64.S:433 The reason is that kvm_put_kvm is causing the destruction of the VM, but the page fault is still on the ->queue list. The ->queue list is owned by the VCPU, not by the work items, so we cannot just add list_del to the work item. Instead, use work->vcpu to note async page faults that have been resolved and will be processed through the done list. There is no need to flush those. Cc: Dmitry Vyukov Signed-off-by: Paolo Bonzini Signed-off-by: Radim Krčmář --- diff --git a/virt/kvm/async_pf.c b/virt/kvm/async_pf.c index 8035cc1eb955..efeceb0a222d 100644 --- a/virt/kvm/async_pf.c +++ b/virt/kvm/async_pf.c @@ -91,6 +91,7 @@ static void async_pf_execute(struct work_struct *work) spin_lock(&vcpu->async_pf.lock); list_add_tail(&apf->link, &vcpu->async_pf.done); + apf->vcpu = NULL; spin_unlock(&vcpu->async_pf.lock); /* @@ -113,6 +114,8 @@ static void async_pf_execute(struct work_struct *work) void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) { + spin_lock(&vcpu->async_pf.lock); + /* cancel outstanding work queue item */ while (!list_empty(&vcpu->async_pf.queue)) { struct kvm_async_pf *work = @@ -120,6 +123,14 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) typeof(*work), queue); list_del(&work->queue); + /* + * We know it's present in vcpu->async_pf.done, do + * nothing here. + */ + if (!work->vcpu) + continue; + + spin_unlock(&vcpu->async_pf.lock); #ifdef CONFIG_KVM_ASYNC_PF_SYNC flush_work(&work->work); #else @@ -129,9 +140,9 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu) kmem_cache_free(async_pf_cache, work); } #endif + spin_lock(&vcpu->async_pf.lock); } - spin_lock(&vcpu->async_pf.lock); while (!list_empty(&vcpu->async_pf.done)) { struct kvm_async_pf *work = list_first_entry(&vcpu->async_pf.done,