[PATCH] config, x86 enable boot cpu hotplug

[Prarit Bhargava]

On 01/20/2014 09:13 AM, Josh Boyer wrote:
> On Sun, Jan 19, 2014 at 07:43:06AM -0500, Prarit Bhargava wrote:
>>
>>
>> On 01/14/2014 08:40 AM, Josh Boyer wrote:
>>> On Tue, Jan 14, 2014 at 7:36 AM, Prarit Bhargava <prarit at redhat.com> wrote:
>>>> Set CONFIG_BOOTPARAM_HOTPLUG_CPU0 to 'y'.  Currently, to disable CPU0,
>>>> the "boot" cpu, one must specify cpu0_hotplug as a kernel parameter.
>>>> Setting this config variable to 1 enables it by default.
>>>>
>>>> I've tested this on systems where it was expected to work and it seems to
>>>> function properly.
>>>
>>> Looking this over, it simply makes it _possible_ to offline CPU0, but
>>> doesn't do it by default, correct?
>>>
>>> I've got it queued locally and assuming the answer to my question is
>>> what I think it is, I'll push it out soonish.  Thanks.
>>>
>>
>> Josh, after some additional testing it looks like we should not do this yet.
>> There are some odd bugs that have popped up on older hardware and a few issues
>> on new hardware that need to be resolved.
>>
>> So ... I think this should be rejected or reverted.
> 
> OK.  Have a link to where these problems cropped up?

Not yet Josh -- I'm still gathering reports.  Here's basically what I have so
far (I'm bogged down with a few things and this isn't critical so it has lower
priority).

We're seeing some large systems that should properly support cpu0 hotplug report
back that they do not.  I'm not sure why, other than I believe that some of the
init code may not properly recognize cpu0 hotplug flags.  I've also seen some
systems that should not have booted with cpu0 hotplug claim to support it ... so
something is broken :( ... I'm just not sure what.

FWIW, I requested a revert in RHEL7 for this, and QA reported that the revert
fixed some other issues with perf.

P.
> 
> josh