[systemd-devel] [PATCH] Drop the udev firmware loader

[Greg KH]

On Fri, May 30, 2014 at 05:00:34AM +0200, Zbigniew Jędrzejewski-Szmek wrote:
> On Thu, May 29, 2014 at 07:51:29PM -0700, Greg KH wrote:
> > On Fri, May 30, 2014 at 04:41:01AM +0200, Michael Biebl wrote:
> > > 2014-05-30 4:32 GMT+02:00 Michael Biebl <mbiebl at gmail.com>:
> > > > 2014-05-30 4:26 GMT+02:00 Greg KH <gregkh at linuxfoundation.org>:
> > > >
> > > >> You update systemd but you don't update the kernel?  How does that make
> > > >> any sense?
> > > >
> > > > There might be very valid reasons why you need to stick with the old
> > > > kernel. As said, one example could be that the new one simply doesn't
> > > > boot. Requiring lock-step upgrades makes the system less
> > > > fault-tolerant.
> > > > So where possible this should be avoided.
> > > 
> > > What I'm trying to say here is: let's rip this code out once all
> > > stable distros out there in the wild ship a kernel with builti-in
> > > firmware loader support, but please not before.
> > 
> > What is "all"?
> > 
> > Do we really have to wait 10+ years just because some random disto
> > doesn't want to update their kernel?
> > 
> > Since when does systemd care about what random distros do?
> It should. I'd hardly call Debian a random distro.

It's a slow moving distro.  If it takes 5 years for their next stable
release, why should we be forced to maintain code that no one else uses?

> They are bound to have lots of issues with the conversion to systemd
> as the default on their wide range of supported systems. Keep such
> compatibility stuff around just makes things easier for everyone. Even
> for us, the cost of maintaining some ifdef-ed code is lower than
> handling the additional bug reports from people who compile their own
> systemd.

Then it could be made into a simple patch for those distros that require
obsolete kernels :)

> > > Keeping the userspace firmware loader for a little longer costs us
> > > nothing and simplifies things for downstreams. So let's be nice to
> > > them. I don't see a good reason to rush the removal of the userspace
> > > firmware loader.
> > 
> > Why not just port the in-kernel firmware loader to those kernel
> > versions?  That solves more problems than sticking with this code
> > does...
> This doesn't readily solve the problem with requiring lock-step
> upgrades.  You first have to deploy the updated kernel, then install
> newer systemd.

You can install a new systemd then say that no more firmware loading
will happen until you reboot with your new kernel that you install
afterwards.  As firmware loading is usually a boot-only event, or a "add
a new device" event, I don't see that as a major restriction, do you?

thanks,

greg k-h