Re: [PATCH v07 2/9] hotplug/cpu: Add operation queuing function

[Michael Ellerman <mpe@ellerman.id.au>]

Michael Bringmann <mwb@linux.vnet.ibm.com> writes:
> On 07/23/2018 10:54 AM, John Allen wrote:
>> On Fri, Jul 13, 2018 at 03:18:01PM -0500, Michael Bringmann wrote:
>>> @@ -407,6 +413,61 @@ void queue_hotplug_event(struct pseries_hp_errorlo=
g *hp_errlog,
...
>>> +
>>> +static int dlpar_pmt(struct pseries_hp_errorlog *work)
>>> +{
>>> +=C2=A0=C2=A0=C2=A0 struct list_head *pos, *q;
>>> +
>>> +=C2=A0=C2=A0=C2=A0 ssleep(15);
>>=20
>> Why do we need to sleep for so long here?
>
> One or more drivers were finishing their initialization about the same
> time as the 'migration_store' operation was completing.  E.g. 'ibmvscsi'
> When we did 'dlpar cpu readd' operations at the same time early on in
> testing, the errors / warnings were inconsistent.  I put in a delay to
> get beyond the driver re-init after migration.  15 seconds was an estimat=
e.
> As it occurs after the completion of the 'migration store' operation,
> it does not delay any responses returned to the HMC.
>
> We can rerun tests with reduced / removed delays to see if they are still
> necessary given some other corrections in the mix.

Please do.

I'm not inclined to merge code with "sleep 15" in it, unless there's
some incredibly good reason for it.

cheers