Re: [Qemu-devel] [RFC v3] translate-all: protect code_gen_buffer with RCU

[Alex Bennée]

Paolo Bonzini <address@hidden> writes:

> On 09/05/2016 13:50, Alex Bennée wrote:
>> > Which locks?  tb_lock during tb_find_fast?  The problem with that was
>> > that it slowed down everything a lot, wasn't it?
>>
>> Very much so, in the new tree (coming soon) with QHT I was able to
>> remove the locks from the whole hot-path which means they where only
>> needed for code generation.
>
> Okay, I'm curious now. :)

https://github.com/stsquad/qemu/commits/mttcg/base-patches-v3 is the
current WIP, with:

https://github.com/stsquad/qemu/commit/0823f1c77f12ed5958f77484d6477ea205aee220

being the commit that clears the hot-path to run without locks.

The tree is based on tcg-next which has made things a lot cleaner now a
bunch of Sergey's stuff has been grabbed by rth. Obviously being WIP
subject to change. Once I'm done with my current out-of-tree diversions
I'll be back onto cleaning the tree up for the next review round.

Review comments on the posted tree's always welcome of course ;-)

>
>> > To me, the RCU idea is not really about making tb_flush (the rare case)
>> > faster; it was more about keeping the rest simple and fast.
>>
>> I'm not sure it achieved that as there is added complexity from having
>> the split buffer and then ensuring you don't double-flush.
>
> Agreed.
>
> Paolo


--
Alex Bennée