Nouveau - Mar 2014 - [PATCH] drm: compute runpm on load, don't register autosuspend for non-runpm

> On Thu, Mar 20, 2014 at 1:28 AM, Ilia Mirkin <imirkin at
alum.mit.edu> wrote:
> > There was a proliferation of duplicated checks for runpm == -1
&&
> > optimus. Instead of continuing that tradition, get rid of all of them,
> > only doing the optimus computation once on load.
> >
> > This should hopefully fix secondary cards suspending and then being
> > unable to come back in non-optimus setups.
> >
> > Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
> > Cc: <stable at vger.kernel.org> # 3.12+
> Also untested, but looks like a saner idea to me.  Dave?
Unfortunately not sane, we are looking at adding apple-gmux support, so the load
ordering is indeterminate, so doing this will mean we just have to undo it
later.

Dave.

On Wed, Mar 19, 2014 at 7:45 PM, David Airlie <airlied at redhat.com>
wrote:
>
>> On Thu, Mar 20, 2014 at 1:28 AM, Ilia Mirkin <imirkin at
alum.mit.edu> wrote:
>> > There was a proliferation of duplicated checks for runpm == -1
&&
>> > optimus. Instead of continuing that tradition, get rid of all of
them,
>> > only doing the optimus computation once on load.
>> >
>> > This should hopefully fix secondary cards suspending and then
being
>> > unable to come back in non-optimus setups.
>> >
>> > Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>> > Cc: <stable at vger.kernel.org> # 3.12+
>> Also untested, but looks like a saner idea to me.  Dave?
>
> Unfortunately not sane, we are looking at adding apple-gmux support, so the
load
> ordering is indeterminate, so doing this will mean we just have to undo it
later.
So... suggestions then? I always end up having to add runpm=0 on
multi-gpu cards... the secondary gpu seems to suspend without any way
back.

----- Original Message -----
> From: "Ilia Mirkin" <imirkin at alum.mit.edu>
> To: "David Airlie" <airlied at redhat.com>
> Cc: "Ben Skeggs" <skeggsb at gmail.com>, "Ben
Skeggs" <bskeggs at redhat.com>, nouveau at lists.freedesktop.org
> Sent: Thursday, 20 March, 2014 9:49:47 AM
> Subject: Re: [Nouveau] [PATCH] drm: compute runpm on load, don't
register autosuspend for non-runpm
> 
> On Wed, Mar 19, 2014 at 7:45 PM, David Airlie <airlied at redhat.com>
wrote:
> >
> >> On Thu, Mar 20, 2014 at 1:28 AM, Ilia Mirkin <imirkin at
alum.mit.edu> wrote:
> >> > There was a proliferation of duplicated checks for runpm ==
-1 &&
> >> > optimus. Instead of continuing that tradition, get rid of all
of them,
> >> > only doing the optimus computation once on load.
> >> >
> >> > This should hopefully fix secondary cards suspending and then
being
> >> > unable to come back in non-optimus setups.
> >> >
> >> > Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
> >> > Cc: <stable at vger.kernel.org> # 3.12+
> >> Also untested, but looks like a saner idea to me.  Dave?
> >
> > Unfortunately not sane, we are looking at adding apple-gmux support,
so the
> > load
> > ordering is indeterminate, so doing this will mean we just have to
undo it
> > later.
> 
> So... suggestions then? I always end up having to add runpm=0 on
> multi-gpu cards... the secondary gpu seems to suspend without any way
> back.
Figure out why you get suspended, send patch to fix it. The conditions should be
enough to block it.

Dave.