Re: [PATCH 0/4] vl: Prioritize device realizations

[David Hildenbrand]

On 21.10.21 10:00, Peter Xu wrote:
> On Thu, Oct 21, 2021 at 09:17:57AM +0200, David Hildenbrand wrote:
>> I know, whenever someone proposes a way to tackle part of a challenging
>> problem, everybody discovers their hopes and dreams and suddenly you
>> have to go all the way to solve the complete problem. The end result is
>> that there is no improvement at all instead of incremental improvement.
> 
> Yeah, there's the trade-off; we either not moving forward or otherwise we 
> could
> potentially bring (more) chaos so the code is less maintainable.  Before I'm
> sure I won't do the latter and convince the others, I need to hold off a bit. 
> :-)

Sure :)

>> I'm not planning on letting the user set the actual number of memslots
>> to use, only an upper limit. But to me, it's fundamentally the same: the
>> user has to enable this behavior explicitly.
> 
> I'm not familiar enough on virtio-mem's side, it's just that it will stop
> working when the ideal value (even in a very corner case) is less than the
> maximum specified, then that trick stops people from specifying the ideal.  
> But
> if it's bigger the better then indeed I don't see much to worry.

Usually it's "the bigger the better", but there are a lot of exceptions,
and error handling on weird user input is a little hairy ... but I'm
playing with it right now, essentially having

"memslots=0" -> auto detect as good as possible
"memslots=1" -> default
"memslits>1" -> use user input, bail out if some conditions aren't met.
Especially, fail plugging if there are not sufficient free memslots around.

-- 
Thanks,

David / dhildenb