From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:52310) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eEqzY-0000Nf-Pq for qemu-devel@nongnu.org; Wed, 15 Nov 2017 01:16:10 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eEqzV-0003mo-Ju for qemu-devel@nongnu.org; Wed, 15 Nov 2017 01:16:08 -0500 Received: from szxga04-in.huawei.com ([45.249.212.190]:2353) by eggs.gnu.org with esmtps (TLS1.0:RSA_ARCFOUR_SHA1:16) (Exim 4.71) (envelope-from ) id 1eEqzV-0003gt-0L for qemu-devel@nongnu.org; Wed, 15 Nov 2017 01:16:05 -0500 From: Yori Fang References: <212a9227-5a56-7a0e-1149-2d1bc884b4f0@huawei.com> Message-ID: Date: Wed, 15 Nov 2017 14:15:23 +0800 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit Subject: Re: [Qemu-devel] QEMU abort when network serivce is restarted during live migration with vhost-user as the network backend List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: =?UTF-8?Q?Marc-Andr=c3=a9_Lureau?= Cc: "qemu-devel@nongnu.org" 在 2017/11/14 19:40, Marc-André Lureau 写道: > Hi > > On Tue, Nov 14, 2017 at 8:09 AM, fangying wrote: >> Hi all, >> >> We have a vm running migration with vhost-user as network backend, we notice that qemu will abort when openvswitch is restarted >> when MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward) is called. The reasion is clear that vhost_dev_set_log returns -1 because >> the network connection is temporarily lost due to the restart of openvswitch service. >> >> Below is the trace of the call stack. >> >> #0 0x00007f868ed971d7 in raise() from /usr/lib64/libc.so.6 >> #1 0x00007f868ed988c8 in abort() from /usr/lib64/libc.so.6 >> #2 0x00000000004d0d35 in vhost_log_global_start (listener=) at /usr/src/debug/qemu-kvm-2.8.1/hw/virtio/vhost.c:794 >> #2 0x0000000000486bd2 in memory_global_dirty_log_start at /usr/src/debug/qemu-kvm-2.8.1/memory.c:2304 >> #3 0x0000000000486dcd in ram_save_init_globals at /usr/src/debug/qemu-kvm-2.8.1/migration/ram.c:2072 >> #4 0x000000000048c185 in ram_save_setup (f=0x25e6ac0, opaque=) at /usr/src/debug/qemu-kvm-2.8.1/migration/ram.c:2093 >> #5 0x00000000004fbee2 in qemu_savevm_state_begin at /usr/src/debug/qemu-kvm-2.8.1/migration/savevm.c:956 >> #6 0x000000000083d8f8 in migration_thread at migration/migration.c:2198 >> >> static void vhost_log_global_start(MemoryListener *listener) >> { >> int r; >> >> r = vhost_migration_log(listener, true); >> if (r < 0) { >> abort(); /* branch taken */ >> } >> } >> >> What confuse me is that >> 1. do we really need to abort here ? > > Not if we have a sane way to handle the situation. It make sense > though to not want to support that use case (restarting the vhost-user > process during migration). > >> 2. all member of callbacks in MemoryListener returned with type void, we cannot judge in any upper function on the call stack. >> Can we just cancel migration here instead of calling abort ? like: > > That would be acceptable to me, but there should be a better way than > calling qmp_migrate_cancel() (we need to give a reason for cancelling, > and report it to user). Juan should be able to help. I agree with you, we'd better give more details here instead of passing NULL in qmp_migrate_cancel. > >> >> diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c >> index ddc42f0..27ae4a2 100644 >> --- a/hw/virtio/vhost.c >> +++ b/hw/virtio/vhost.c >> @@ -27,6 +27,7 @@ >> #include "hw/virtio/virtio-access.h" >> #include "migration/blocker.h" >> #include "sysemu/dma.h" >> +#include "qmp-commands.h" >> >> /* enabled until disconnected backend stabilizes */ >> #define _VHOST_DEBUG 1 >> @@ -885,7 +886,7 @@ static void vhost_log_global_start(MemoryListener *listener) >> >> r = vhost_migration_log(listener, true); >> if (r < 0) { >> - abort(); >> + qmp_migrate_cancel(NULL); >> } >> } >> >> @@ -895,7 +896,7 @@ static void vhost_log_global_stop(MemoryListener *listener) >> >> r = vhost_migration_log(listener, false); >> if (r < 0) { >> - abort(); >> + qmp_migrate_cancel(NULL); >> } >> } >> >> >> >> > > >