On 10/17/2017 02:36 PM, Thomas Huth wrote:
> On 17.10.2017 18:27, Daniel Henrique Barboza wrote:
>>
>> On 10/17/2017 11:32 AM, Thomas Huth wrote:
>>> So far the CPU hot-plug qtest was only checking "cpu-add" on x86. With
>>> these patches, we now test "device_add" for hot-plugging CPUs on x86,
>>> and enable the test on ppc64 and s390x, too.
>> Question: is there a reason other than "no one bothered" to not have a
>> CPU hot
>> unplug test, something around the lines of the now cpu-plug-test.c? I might
>> give it a shot if there is no known roadblocks against it.
> AFAIK on x86, you need some ACPI magic on the guest side to do this
> (something similar to qpci_unplug_acpi_device_test() in
> tests/libqos/pci-pc.c I guess). And on s390x, CPU unplug is not
> supported at all. But it might work on ppc64 without guest intervention,
> not sure - maybe Bharata can comment on this...
Good point, it won't work in spapr machine without guest intervention too.
Both CPU and LMB unplug relies on guest-side callbacks to complete the
operation.
I'll see how this ACPI magic you mentioned is done and see how hard it is to
do it for ppc64.
Daniel
>
> Thomas
>