On Thu, Jan 10, 2013 at 8:18 PM, Darren Hart <[email protected]> wrote:

>
>
> On 01/10/2013 05:06 PM, Liang Li wrote:
> > On 2013-01-11 05:24, Bruce Ashfield <[email protected]>
> wrote:
> >> On 13-01-08 02:18 AM, Liang Li wrote:
> >>> On 2013-01-08 09:14, Darren Hart <[email protected]> wrote:
> >>>>
> >>>>
> >>>> On 01/07/2013 05:11 PM, Liang Li wrote:
> >>>>> On 2013-01-08 08:59, Liang Li <[email protected]> wrote:
> >>>>>> On 2013-01-08 01:00, Darren Hart <[email protected]> wrote:
> >>>>>>> Hrm, is this right with respect to the KMS driver? If we specify a
> >>>>>>> console resolution requiring KMS, does this work with DRM_I915 as
> a module?
> >>>>>>>
> >>>>>>
> >>>>>> Yes, I think so. :)
> >>>>>
> >>>>> Er, I mean, boot progress would be fine .. just delayed KMS
> >>>>> initialization then console resolution would be changed after the
> >>>>> module loaded.
> >>>>
> >>>> Personally I'd prefer this stayed as a =y. I see very little need for
> >>>
> >>> Actually, I see 'm' will be more flexible. Eg, when its be 'y',
> >>> intel-agp also *have to be* 'y'. But with a 'm', it would offer
> >>> convenience.
> >>>
> >>>> modules in kernels tailored for specific systems. i915 config is
> already
> >>>> separated out... why make KMS a module?
> >>>>
> >>>
> >>> Not for the kernel configure option, its still be 'y', but as soon as
> >>> drm-i915 be 'm', its actually be 'm'. :)
> >>>
> >>> Besides, I see distroes make drm drivers be modules as much as
> >>> possible, seems it would works good, reduced kernel footprint, gain
> >>> flexible etc. So I wonder if you know some explicit benefits that I
> >>> did not aware of .. to support it stay as 'y' ? :)
> >>
> >> bump. I'm sitting on all these patches until the Intel folks do an
> >> ack / nack or otherwise on the queue .. since they are the creators
> >> of the configs!
> >>
> >
> > Yes, I am waiting to hear response as well, since I think Darren or
> > other folks might a bit busy. But wait one or two more days seems ok
> > to me. :) Hopefully I'll get ack.
>
> Sorry guys, avalanche of new work piling up over the stale pre-vacation
> work.
>
>
No problem at all. I was just trying to get a few email out of my inbox at
the end
of the day and figured I'd ping.

I should have pointed out in my ping that don't assume I'm for or against
the
series as is stands, I'm the one that asked Liang to send it along, since I
wanted to gather more thoughts before weighing in.


> 1) I don't think what the distros do is particularly relevant as they
>    have very different goals. They use modules because it allows them to
>    make more generic images. You don't want to bloat your kernel with
>    features only some systems will use. In our case we build
> system-specific
>    kernels where very little of what is configured in will not be used.
>

Agreed. Tight configs should always be a top goal, and I agree, built in is
my
preference to a pile of modules floating around a disk, since we know our
configs, create a baseline and encourage anyone inheriting them to tweak to
their taste.


>
> 2) I remain strongly opposed to modules. If we are going to make such a
>    change, I would want to see real numbers demonstrating WHY it is a good
>    idea. How much space do we actually save? What does that space buy us in
>    boot time?
>

Also agreed.


>
> 3) The impact on users of the existing fragments needs to be well
>    understood, "Yes, I think so. :-)" really isn't enough.
>
>
This was also my concern as well, there are existing users and use cases, we
need to be careful changing the configs once they've been in tree for a
while.

I'm good with 2/3 and 3/3 of the series, they are legitimate bug fixes. As
for
changing I915 to a module, I'd suggest we leave it as-is in the base yocto
configuration and it can be tweaked to 'y' in the Wind River kernel ..
using a
feature fragment of a different name (since we want to consume the yocto
baseline as-is).

Cheers,

Bruce


> So that's my position on making platform specific support options into
> modules.
>
> --
> Darren Hart
> Intel Open Source Technology Center
> Yocto Project - Technical Lead - Linux Kernel
> _______________________________________________
> linux-yocto mailing list
> [email protected]
> https://lists.yoctoproject.org/listinfo/linux-yocto
>



-- 
"Thou shalt not follow the NULL pointer, for chaos and madness await thee
at its end"
_______________________________________________
linux-yocto mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/linux-yocto

Reply via email to