From: XuYandong <xuyandong2@huawei.com>
After vcpu1 thread exiting, vcpu0 thread (received notification) is still waiting for
holding qemu_global_mutex in cpu_remove_sync, at this moment, vcpu1 is still in global cpus list.
If main thread grab qemu_global_mutex in order to handle qmp command "info cpus",
qmp_query_cpus visit unpluged vcpu1 will lead qemu process to exit.
Signed-off-by: XuYandong <xuyandong2@huawei.com>
---
cpus.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/cpus.c b/cpus.c
index 2cb0af9..9b3a6c4 100644
--- a/cpus.c
+++ b/cpus.c
@@ -2018,6 +2018,11 @@ CpuInfoList *qmp_query_cpus(Error **errp)
CPU_FOREACH(cpu) {
CpuInfoList *info;
+
+ if (cpu->unplug) {
+ continue;
+ }
+
#if defined(TARGET_I386)
X86CPU *x86_cpu = X86_CPU(cpu);
CPUX86State *env = &x86_cpu->env;
--
1.8.3.1
On Wed, 11 Apr 2018 19:16:02 +0800 linzhecheng <linzhecheng@huawei.com> wrote: > From: XuYandong <xuyandong2@huawei.com> > > After vcpu1 thread exiting, vcpu0 thread (received notification) is still waiting for > holding qemu_global_mutex in cpu_remove_sync, at this moment, vcpu1 is still in global cpus list. > If main thread grab qemu_global_mutex in order to handle qmp command "info cpus", > qmp_query_cpus visit unpluged vcpu1 will lead qemu process to exit. Add here exact error or better stack trace in case it crashes. > Signed-off-by: XuYandong <xuyandong2@huawei.com> > --- > cpus.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/cpus.c b/cpus.c > index 2cb0af9..9b3a6c4 100644 > --- a/cpus.c > +++ b/cpus.c > @@ -2018,6 +2018,11 @@ CpuInfoList *qmp_query_cpus(Error **errp) > > CPU_FOREACH(cpu) { > CpuInfoList *info; > + > + if (cpu->unplug) { > + continue; > + } Shouldn't be it done for qmp_query_cpus_fast() as well? > + > #if defined(TARGET_I386) > X86CPU *x86_cpu = X86_CPU(cpu); > CPUX86State *env = &x86_cpu->env;
Yes, we should add it at qmp_query_cpus_fast() . In this case, qmp_query_cpus() may kick a vcpu thread that has stopped causing the qemu process to exit, qmp_query_cpus_fast() does not interrupt vCPU threads, but may return invalid vCPU information. > -----Original Message----- > From: Igor Mammedov [mailto:imammedo@redhat.com] > Sent: Wednesday, April 11, 2018 7:46 PM > To: linzhecheng <linzhecheng@huawei.com> > Cc: qemu-devel@nongnu.org; xuyandong <xuyandong2@huawei.com>; > pbonzini@redhat.com; wangxin (U) <wangxinxin.wang@huawei.com>; > rth@twiddle.net; crosthwaite.peter@gmail.com > Subject: Re: [Qemu-devel] [PATCH] cpu: skip unpluged cpu when querying > cpus > > On Wed, 11 Apr 2018 19:16:02 +0800 > linzhecheng <linzhecheng@huawei.com> wrote: > > > From: XuYandong <xuyandong2@huawei.com> > > > > After vcpu1 thread exiting, vcpu0 thread (received notification) is > > still waiting for holding qemu_global_mutex in cpu_remove_sync, at this > moment, vcpu1 is still in global cpus list. > > If main thread grab qemu_global_mutex in order to handle qmp command > > "info cpus", qmp_query_cpus visit unpluged vcpu1 will lead qemu process > to exit. > Add here exact error or better stack trace in case it crashes. > > > > Signed-off-by: XuYandong <xuyandong2@huawei.com> > > --- > > cpus.c | 5 +++++ > > 1 file changed, 5 insertions(+) > > > > diff --git a/cpus.c b/cpus.c > > index 2cb0af9..9b3a6c4 100644 > > --- a/cpus.c > > +++ b/cpus.c > > @@ -2018,6 +2018,11 @@ CpuInfoList *qmp_query_cpus(Error **errp) > > > > CPU_FOREACH(cpu) { > > CpuInfoList *info; > > + > > + if (cpu->unplug) { > > + continue; > > + } > Shouldn't be it done for qmp_query_cpus_fast() as well? > > > + > > #if defined(TARGET_I386) > > X86CPU *x86_cpu = X86_CPU(cpu); > > CPUX86State *env = &x86_cpu->env;
© 2016 - 2024 Red Hat, Inc.