A few notes on testing

Matthew Burgess matthew at linuxfromscratch.org
Wed Jul 14 09:43:55 PDT 2004


On Wed, 14 Jul 2004 03:03:58 +0100
Ian Molton <spyro at f2s.com> wrote:

> On Tue, 13 Jul 2004 19:20:06 +0100
> Matthew Burgess <matthew at linuxfromscratch.org> wrote:
> 
> > 
> > Then I wonder how hotplug will do it's coldplugging then? 
> 
> stuff like that is what syusfs is for (among other things).
> basuically sysfs is where stuff in /proc *should* go if its not
> process information.

Yes, but my whole question was about how coldplugging works when /proc
only contains process information and knows nothing about hardware
devices.  At the moment, hotplug .rc scripts scan /proc (and also
/sys) to detect what's currently attached to a device.  Now, when /proc
goes into 'process information' mode only, then we're reliant on /sys. 
But /sys is only populated when the device driver is loaded.  If the
device driver is a module then /sys knows nothing about it and udev
can't create the device because it needs sysfs info to do so.  In order
for the module to populate /sys it needs to be loaded, but it can't do
so because the device can't be accessed (it doesn't exist yet) and we go
round and round in circles.  Now, as Kevin explained, as long as
the *bus* drivers are loaded (which implies they should *never* be
modules) then this should allow the necessary probing (via
/sys/bus/[bus]) to occur.

Cheers,

Matt.



More information about the lfs-dev mailing list