From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:46275) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1daykZ-0002wo-Ub for qemu-devel@nongnu.org; Fri, 28 Jul 2017 02:27:52 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1daykW-0005zk-Gk for qemu-devel@nongnu.org; Fri, 28 Jul 2017 02:27:51 -0400 Received: from szxga02-in.huawei.com ([45.249.212.188]:4407) by eggs.gnu.org with esmtps (TLS1.0:RSA_ARCFOUR_SHA1:16) (Exim 4.71) (envelope-from ) id 1daykV-0005uL-TI for qemu-devel@nongnu.org; Fri, 28 Jul 2017 02:27:48 -0400 References: <1501122974-18860-1-git-send-email-jianjay.zhou@huawei.com> <5c0789ef-9c29-b867-5e84-8624d04c16cc@redhat.com> From: Jay Zhou Message-ID: <597AD8F3.3020702@huawei.com> Date: Fri, 28 Jul 2017 14:25:55 +0800 MIME-Version: 1.0 In-Reply-To: <5c0789ef-9c29-b867-5e84-8624d04c16cc@redhat.com> Content-Type: text/plain; charset="utf-8"; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH v2] migration: optimize the downtime List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Paolo Bonzini , qemu-devel@nongnu.org Cc: dgilbert@redhat.com, arei.gonglei@huawei.com, zhang.zhanghailiang@huawei.com, wangxinxin.wang@huawei.com, weidong.huang@huawei.com, xiaoguangrong@tencent.com, jdenemar@redhat.com, huangzhichao@huawei.com Hi Paolo, On 2017/7/27 22:15, Paolo Bonzini wrote: > On 27/07/2017 04:36, Jay Zhou wrote: >> Qemu_savevm_state_cleanup takes about 300ms in my ram migration tests >> with a 8U24G vm(20G is really occupied), the main cost comes from >> KVM_SET_USER_MEMORY_REGION ioctl when mem.memory_size = 0 in >> kvm_set_user_memory_region. In kmod, the main cost is >> kvm_zap_obsolete_pages, which traverses the active_mmu_pages list to >> zap the unsync sptes. >> >> It can be optimized by delaying memory_global_dirty_log_stop to the next >> vm_start. >> >> Changes v1->v2: >> - create a VMChangeStateHandler in memory.c to reduce the coupling [Paolo] >> >> Signed-off-by: Jay Zhou > > memory_vm_change_state_handler should delete the handler. > > Apart from that, because there is no protection against nested > invocations of memory_global_dirty_log_start/stop, a little more work is > needed. Thank for your patience and pointing everything out, I will fix all of them. Jay >> --- >> memory.c | 27 ++++++++++++++++++++++++++- >> 1 file changed, 26 insertions(+), 1 deletion(-) >> >> diff --git a/memory.c b/memory.c >> index a7bc70a..4c22b7e 100644 >> --- a/memory.c >> +++ b/memory.c >> @@ -2357,8 +2357,14 @@ void memory_global_dirty_log_sync(void) >> } >> } >> >> +static VMChangeStateEntry *vmstate_change; >> + >> void memory_global_dirty_log_start(void) >> { >> + if (vmstate_change) { >> + qemu_del_vm_change_state_handler(vmstate_change); > > This should also NULL vmstate_change, so that you can detect the case > where the handler is already installed. > >> + } >> + >> global_dirty_log = true; >> >> MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward); >> @@ -2369,7 +2375,7 @@ void memory_global_dirty_log_start(void) >> memory_region_transaction_commit(); >> } >> >> -void memory_global_dirty_log_stop(void) >> +static void memory_global_dirty_log_do_stop(void) >> { >> global_dirty_log = false; >> >> @@ -2381,6 +2387,25 @@ void memory_global_dirty_log_stop(void) >> MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse); >> } >> >> +static void memory_vm_change_state_handler(void *opaque, int running, >> + RunState state) >> +{ >> + if (running) { >> + memory_global_dirty_log_do_stop(); > > Because this will delete the handler, it should also NULL vmstate_change. > >> + } >> +} >> + >> +void memory_global_dirty_log_stop(void) >> +{ >> + if (!runstate_is_running()) { >> + vmstate_change = qemu_add_vm_change_state_handler( >> + memory_vm_change_state_handler, NULL); > > And this needs to exit immediately if you have an installed handler. > > Thanks, > > Paolo > >> + return; >> + } >> + >> + memory_global_dirty_log_do_stop(); >> +} >> + >> static void listener_add_address_space(MemoryListener *listener, >> AddressSpace *as) >> { >> > > > . >