[PATCH v1 02/10] qemu_domain.c: align memory modules before calculating 'initialmem'

[Daniel Henrique Barboza]

On 11/18/20 4:44 AM, Peter Krempa wrote:
> On Mon, Nov 16, 2020 at 20:43:09 +0100, Andrea Bolognani wrote:
>> On Fri, 2020-11-13 at 16:23 -0300, Daniel Henrique Barboza wrote:
>>> On 11/13/20 10:51 AM, Andrea Bolognani wrote:
>>>> I only skimmed the remaining patches and didn't dig into the code as

[...]


>>
>> In general, changing guest ABI between boots is not something that we
>> want to happen.
>>
>> I have trouble keeping all the details of memory alignment inside my
>> head and I can't quite spend the time necessary to swap them back in
>> right now, so please apologies if I'm being vague and of course
>> correct me if I'm wrong... Having Peter in the thread will also
>> surely help with that :)
> 
> Either you have high hopes or my sarcasm detector is failing.
> 
> Just as one data point: the 'PostParse' callback happen _always_. Even
> when migrating the VM. My patch according to the commit message is
> specifically avoiding the alignment on migration. Thus the code can't be
> moved to the post parse callback.

Not without considering the PARSE_ABI_UPDATE flag, and even then I'm not
sure if it's worth the trouble like I mentioned earlier.

> 
> The outcome documented in the NEWS just mentions that it's to update the
> live XML.
> 
> Neither of the commit messages of this series mentions that there is an
> issue with migration so the series needs to be re-evaluated in that
> light.
> 

This is correct. I got a lot of stuff wrong in the v1. v2 will be way
shorter in scope but it will not mess up with existing migration/snapshot
mechanics. My aim will be to fix the existing alignment calculation for
pseries DIMMs.


Thanks,


DHB