Re: [kvm-unit-tests PATCH v6 07/10] s390x: css: msch, enable test

[Pierre Morel <pmorel@linux.ibm.com>]

On 2020-05-15 10:53, Janosch Frank wrote:
> On 5/15/20 10:25 AM, Cornelia Huck wrote:
>> On Fri, 15 May 2020 09:11:52 +0200
>> Pierre Morel <pmorel@linux.ibm.com> wrote:
>>
>>> On 2020-05-14 14:08, Cornelia Huck wrote:
>>>> On Tue, 28 Apr 2020 10:27:36 +0200
>>>> Pierre Morel <pmorel@linux.ibm.com> wrote:
>>>>    
>>>>> On 2020-04-27 15:11, Janosch Frank wrote:
>>>>>> On 4/24/20 12:45 PM, Pierre Morel wrote:
>>>>    
>>>>>>> This is NOT a routine to really enable the channel, no retry is done,
>>>>>>> in case of error, a report is made.
>>>>>>
>>>>>> Would we expect needing retries for the pong device?
>>>>>
>>>>> Yes it can be that we need to retry some instructions if we want them to
>>>>> succeed.
>>>>> This is the case for example if we develop a driver for an operating system.
>>>>> When working with firmware, sometime, things do not work at the first
>>>>> time. Mostly due to races in silicium, firmware or hypervisor or between
>>>>> them all.
>>>>>
>>>>> Since our purpose is to detect such problems we do not retry
>>>>> instructions but report the error.
>>>>>
>>>>> If we detect such problem we may in the future enhance the tests.
>>>>
>>>> I think I've seen retries needed on z/VM in the past; do you know if
>>>> that still happens?
>>>>    
>>>
>>> I did not try the tests under z/VM, nor direct on an LPAR, only under
>>> QEMU/KVM.
>>> Under QEMU/KVM, I did not encounter any need for retry, 100% of the
>>> enabled succeeded on first try.
>>
>> Yep, QEMU/KVM should be fine. Do you plan to run this on anything else?
>>
> 
> I'd like to have it compatible with z/VM / LPAR as well if it isn't too
> much work. You never know when you need it and having tests for all
> hypervisors has been quite a help in the past.
> 

OK, then I can add some retries in prevision.



-- 
Pierre Morel
IBM Lab Boeblingen