Re: [Qemu-devel] [RFC v2 08/11] tcg: add options for enabling MTTCG

[Sergey Fedorov]

On 12/04/16 15:50, KONRAD Frederic wrote:
>
>
> Le 12/04/2016 14:42, Sergey Fedorov a écrit :
>> On 12/04/16 14:59, Peter Maydell wrote:
>>> On 12 April 2016 at 12:48, Alex Bennée <address@hidden> wrote:
>>>> Sergey Fedorov <address@hidden> writes:
>>>>> Maybe we'd better use existing qemu accelerator framework and
>>>>> introduce
>>>>> "-machine accel=mttcg" option?
>>>> My worry would be breaking existing code which assumes kvm | tcg. We
>>>> will want to enable mttcg by default for combos that work without
>>>> having
>>>> to tweak tooling that already uses the -machine options.
>>> Comments from the peanut gallery:
>>>   * We definitely want to just default to MTTCG where it works:
>>> users shouldn't have to add new command line switches to get the
>>> better performance etc, because in practice that won't happen
>>>   * Anything that adds a new command line option at all has to
>>> be supported practically forever, so options that are only
>>> useful for a transitional period are worth thinking twice about
>>
>> Yes, but users may like to have an option to disable MTTCG for some
>> reason. I'm also concerned about icount mode: not sure how to account
>> virtual time when all vCPUs run in parallel.
>>
>

>
> I'm thinking the same, we don't have a solution for icount yet.
> The reverse execution support is probably badly broken as well.

Reverse execution doesn't even seem to support more than a single core.
As of icount, it looks like to be incompatible with MTTCG.

Kind regards,
Sergey