Bug#563003: xserver-xorg-video-intel: New KMS method in experimental is not upstream recommended
On Wed, Dec 30, 2009 at 14:00:51 -0600, Jason Clinton wrote:
> On Tue, Dec 29, 2009 at 5:15 PM, Julien Cristau <jcristau@debian.org> wrote:
> > I'm not sure why you're mentioning initramfs-tools, since there seems to
> > be wide agreement that the kms driver shouldn't be loaded in the
> > initramfs.
>
> I'm uncertain what you mean by "wide agreement" because both (Fedora
> and Ubuntu) distributions shipping with KMS by default are loading the
> KMS driver *in* their respective initramfs's. Perhaps you a referring
> to wide agreement among Debian developers?
>
No. I mean a wide agreement in general that the initramfs's job is to
mount the root filesystem and switch to that, not to do random other
stuff.
> > I'm also puzzled by your assertion that this method is "not upstream
> > recommended", since it was acked by Carl Worth (see #555906).
>
> I read his acknowledgement as a blessing on both your packaging and
> the approach of using a modprobe conf. file. And while Carl is
> certainly respected, he is *not* the upstream; that would be Eric
> Anholt and Keith Packard.
Carl is xf86-video-intel's maintainer, not sure what more you need…
Still no evidence for your assertion that this approach isn't
recommended.
> Regardless, your method is not necessarily
> incompatible with what I suggested. Indeed, an initramfs system which
> does copy in the modprobe configuration file is quite compatible with
> your method. It just doesn't work in Debian because that is not what
> was added to initramfs-tools and it's unnecessary in Fedora and Ubuntu
> because of the kernel module parameter that you pointed to. If
> initramfs-tools were extended to copy-in and then honor the modprobe
> configuration data, then "video=i915" would be sufficient for invoking
> KMS (that is, "video=i915:modeset=1" would be unneeded.)
>
The modprobe configuration data *is* copied and honored by
initramfs-tools.
> I'm not saying that this would be the preferred method (indeed you are
> still having to modify the kernel line) but it would be one way. IMHO,
> "video=i915:modeset=1" is the most sane method as long as the kernel
> parameter is to remain as it is currently in Debian.
>
> Either way, it *must* be loaded in the initramfs to preserve VT1 which
> is what this bug is about.
>
That can also be fixed by udev loading i915 when it loads the rest of the
drivers at the start of the boot. No need to change the kernel command
line then.
Cheers,
Julien
Reply to: