udev closure

Alexander E. Patrakov see at the.sig
Sun May 23 09:01:17 PDT 2004


Matthew Burgess wrote:
> On Sun, 23 May 2004 19:12:42 +0600
> "Alexander E. Patrakov" <see at the.sig> wrote:
> 
> 
>>Matthew Burgess wrote:
>>
>>>udev & hotplug will both go into the 6.0 book.
>>
>>Should I do the CVS commit to b6_0 myself? If so, I will do that in 3
>>hours.
> 
> 
> Yes please.

Done.

There are still some things to do for proper support of modular kernels. 
The most important is inclusion of the "modules" script. On IRC, there 
was a conversation between Nathan, Zack and me with the following 
conclusions:

1) The "modules" script should run before mountfs
2) The check for writeable modules.dep is bogus because of that and 
should be omitted
3) I will be notified upon progress by private E-Mail, but there is 
nothing yet.

Also, there is the following lack of functionality in the "modules" 
script. Udev does not work well if any of character or block devices are 
modules (a classical chicken-and-egg problem). If all such devices are 
modules, they will be too much for the "modules" script. The 
LFS-specific solution must be one of those:

A) Document this and don't support kernels with character or block 
devices compiled as modules (IMHO bad)
B) Add this to one of the initscripts:

# WARNING: rejected by hotplug developers
KERNELVERSION=`uname -r`
MODLIST="`cat /lib/modules/$KERNELVERSION/modules.alias \
/etc/modprobe.conf | egrep ' (char|block)-major' \
| awk '{print $3}' | sort -u`"
for MODULE in $MODLIST
do
	modprobe $MODULE
done

C) Invent some other mechanism for loading entire classes of modules.

-- 
Alexander E. Patrakov
To get my address: echo '0!42!+/6 at 5-3.535.25' | tr [!-:] [a-z] | tr n .



More information about the lfs-dev mailing list