On Fri, 8 Jul 2016 15:19:58 +1000
David Gibson <da...@gibson.dropbear.id.au> wrote:
> On Thu, Jul 07, 2016 at 08:20:22PM +0530, Bharata B Rao wrote:
> > Add CPUState::stable_cpu_id and use that as instance_id in
> > vmstate_register() call.
> >
> > Introduce has-stable_cpu_id property that allows target machines to
> > optionally switch to using stable_cpu_id instead of cpu_index.
> > This will help allow successful migration in cases where holes are
> > introduced in cpu_index range after CPU hot removals.
> >
> > Suggested-by: Igor Mammedov <imamm...@redhat.com>
> > Signed-off-by: Bharata B Rao <bhar...@linux.vnet.ibm.com>
> > ---
> > exec.c | 6 ++++--
> > include/qom/cpu.h | 5 +++++
> > qom/cpu.c | 6 ++++++
> > 3 files changed, 15 insertions(+), 2 deletions(-)
> >
> > diff --git a/exec.c b/exec.c
> > index fb73910..3b36fe5 100644
> > --- a/exec.c
> > +++ b/exec.c
> > @@ -619,12 +619,14 @@ static void cpu_release_index(CPUState *cpu)
> > void cpu_vmstate_register(CPUState *cpu)
> > {
> > CPUClass *cc = CPU_GET_CLASS(cpu);
> > + int instance_id = cpu->has_stable_cpu_id ? cpu->stable_cpu_id :
> > + cpu->cpu_index;
> >
> > if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
> > - vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu);
> > + vmstate_register(NULL, instance_id, &vmstate_cpu_common, cpu);
> > }
> > if (cc->vmsd != NULL) {
> > - vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu);
> > + vmstate_register(NULL, instance_id, cc->vmsd, cpu);
> > }
> > }
> >
> > diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> > index 331386f..527c021 100644
> > --- a/include/qom/cpu.h
> > +++ b/include/qom/cpu.h
> > @@ -273,6 +273,9 @@ struct qemu_work_item {
> > * @kvm_fd: vCPU file descriptor for KVM.
> > * @work_mutex: Lock to prevent multiple access to queued_work_*.
> > * @queued_work_first: First asynchronous work pending.
> > + * @stable_cpu_id: Use as instance_id argument in cpu vmstate_register
> > calls
> > + * @has_stable_cpu_id: Set to enforce the use of @stable_cpu_id
> > + * over cpu_index during vmstate registration.
> > *
> > * State of one CPU core or thread.
> > */
> > @@ -360,6 +363,8 @@ struct CPUState {
> > (absolute value) offset as small as possible. This reduces code
> > size, especially for hosts without large memory offsets. */
> > uint32_t tcg_exit_req;
> > + int stable_cpu_id;
> > + bool has_stable_cpu_id;
> > };
> >
> > QTAILQ_HEAD(CPUTailQ, CPUState);
> > diff --git a/qom/cpu.c b/qom/cpu.c
> > index 1095ea1..bae1bf7 100644
> > --- a/qom/cpu.c
> > +++ b/qom/cpu.c
> > @@ -363,6 +363,11 @@ static int64_t cpu_common_get_arch_id(CPUState *cpu)
> > return cpu->cpu_index;
> > }
> >
> > +static Property cpu_common_properties[] = {
> > + DEFINE_PROP_BOOL("has-stable-cpu-id", CPUState, has_stable_cpu_id,
> > false),
>
> It seems odd to me that stable_cpu_id itself isn't exposed as a
> property. Even if we don't need to set it externally for now, it
> really should be QOM introspectable.
Should it? Why?
It's QEMU internal detail and outside world preferably shouldn't
know anything about it.
As example look at cpu_index which were/is used in HMP and -numa
interfaces and now mgmt tries make some sense of it.
Cleaner way should be teaching -numa to handle cpus specified by
socket/core/thread IDs so that mgmt would actually know what CPUs
it assigns to what nodes and not to look at/invent/generate some
internal cpu_id.
>
> > + DEFINE_PROP_END_OF_LIST()
> > +};
> > +
> > static void cpu_class_init(ObjectClass *klass, void *data)
> > {
> > DeviceClass *dc = DEVICE_CLASS(klass);
> > @@ -394,6 +399,7 @@ static void cpu_class_init(ObjectClass *klass, void
> > *data)
> > * IRQs, adding reset handlers, halting non-first CPUs, ...
> > */
> > dc->cannot_instantiate_with_device_add_yet = true;
> > + dc->props = cpu_common_properties;
> > }
> >
> > static const TypeInfo cpu_type_info = {
>
