[Mesa-dev] [PATCH 0/4] Some prep work for fixing multi context issues in Nouveau

Karol Herbst kherbst at redhat.com
Tue Nov 20 22:28:06 UTC 2018


On Tue, Nov 20, 2018 at 10:34 PM Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>
> On Tue, Nov 20, 2018 at 4:12 PM Karol Herbst <kherbst at redhat.com> wrote:
> >
> > On Tue, Nov 20, 2018 at 8:42 PM Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> > >
> > > On Tue, Nov 20, 2018 at 2:22 PM Karol Herbst <kherbst at redhat.com> wrote:
> > > >
> > > > This series cleans up some code in preparation for the real fix and contains
> > > > cleanups we want to have regardless.
> > > >
> > > > The approach in soon to follow patches is to give each contexts its own
> > > > nouveau_client, nouveau_pushbuf and fence list and have operations triggered
> > > > through a context only use objects owned by the context.
> > >
> > > Didn't I already say, many many times, that such an approach was a non-starter?
> > >
> > >   -ilia
> >
> > I don't see why if each thread has it's own pushbuffer internally and
> > doesn't touch others.
>
> Er, perhaps I wasn't specific enough. Each thread (and thus
> nouveau_context) having its own pushbuf is a good idea. Should do
> that. It doesn't solve anything directly, but it simplifies certain
> scenarios.
>
> Each nouveau_context having its own timeline (fence, client, etc) -
> bad idea. Shouldn't do that.
>

why though? I was talking with Ben about it and he actually suggested
the idea about each context having its own client. I am not quite sure
about the fencing stuff though. Right now it seems to work much better
than what we have right now, but I wasn't trying hard enough to check
if it still breaks somewhere. I could fix all the issues I was running
into for now.

>   -ilia


More information about the mesa-dev mailing list