qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 02/10] s390x/cpumodel: remove CSSKE from base mo


From: David Hildenbrand
Subject: Re: [Qemu-devel] [PATCH 02/10] s390x/cpumodel: remove CSSKE from base model
Date: Thu, 18 Apr 2019 15:15:50 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.6.1

On 18.04.19 15:00, Christian Borntraeger wrote:
> 
> 
> On 18.04.19 14:45, David Hildenbrand wrote:
>> On 18.04.19 13:31, Christian Borntraeger wrote:
>>> conditional sske is deprecated and a distant future machine (will be one
>>> where the IBC will not allow to fully go back to z14) will remove this
>>> feature. To prepare for this and allow for the z14 and older cpu model
>>> to still run on systems without csske, remove csske from the base (and
>>
>> will csske feature be a default feature for zNext? Or is it not
>> available *at all*.
>>
>> In case it is not available, baselining and cpu model comparison have to
>> be thought about "ignoring csske".
>>
>>> thus the default models for z10..z14). For compat machines we have to
>>> add those back.
>>
>> Base models are machine-independent. That means, changing base models is
>> not supported.
> 
> Why is that? for the expansion?

Yes, and baselining. It's what the QAPI interface promises.

> 
>> Once we introduce new models like here, we can set the
>> new base models into stone.
> 
> the new model is easy (and yes I could only disable CSSKE in the base
> model for gen15 but not for gen14. 
> 
> The problem is that without some kind of fixup for older base models like
> z10-z14 expansion will fall back to z9 on anything that no longer has
> csske. 

I'd prefer to do generate the "right" base models right away and not fix
them up at runtime. They should never change, so fixing them up at
runtime feels wrong.

>> Luckily, we already have "ignored_base_feat", maybe we can simply add
>> csske there and have it working.
> 
> Yes, maybe we could ignore csske. 
> 

Yes, that should at least avoid falling back to a z9.

-- 

Thanks,

David / dhildenb



reply via email to

[Prev in Thread] Current Thread [Next in Thread]