Re: [Qemu-devel] [PATCH qemu v7 06/14] spapr_iommu: Introduce "enabled" state for TCE table

[Alexey Kardashevskiy <aik@ozlabs.ru>]

On 05/27/2015 01:51 AM, Paolo Bonzini wrote:
>
>
> On 26/05/2015 17:49, Alexey Kardashevskiy wrote:
>>>
>>> It's very little used, but that's just because it's not too common.
>>> There's nothing wrong with it. :)
>>>
>>> If you do del/set_size/add, you may want to put a
>>> memory_region_transaction_{begin,commit} around the whole dance.


>> Here I lost you again :)
>> Why? These are IOMMU MRs -> they are dynamic, what will begin()/commit()
>> change here?
>
> If you don't add them, the memory core may create two or three different
> flatviews.  With begin/commit, it will only do one change.  It's just an
> optimization.



One step back :) Whole dance is what here? There are:
1) del+set_size(0)
2) set_size(not zero)+add

1) and 2) are called in different places, between those the guest get 
control so I cannot wrap these in begin/commit.  If you suggest 
begin+commit around, for example, 1)- this is just a single call and there 
be one change to flatview, where many changes may happen? And only the last 
one is used, right, or there is a list of them somewhere? :)



-- 
Alexey