modutils regression: rmmod wants the module to be installed

[Alessandro Rubini]

Hello.

First of all I apologize for not having investigated this further as
I expect the problem to be trivial to identify and fix for bb developers.

I'm runnign a system with custom modules. On industrial systems I
never "make modules_install" as I don't want the plethora of
directories to just load some custom modules. I just load the
filenames and unload the module names without /lib/modules even
existing.

With 1.13.1, "insmod ./localmod.ko" works as usual, but "rmmod
localmod" doesn't work unless I have /lib/modules/$(uname -r) in the
filesystem. And even in that case it reports failure even though it succeeds.
This seems related to automatic creation of modules.dep.bb .

Example (ARM AT91SAM9260, busybox-only NFS-Root filesystem):

   # insmod ./fiq-task.ko period=20
   # grep '^fiq.task' /proc/modules
   fiq_task 2284 0 - Live 0xbf011000
   # rmmod fiq_task
   rmmod: chdir(/lib/modules): No such file or directory
   # mkdir /lib/modules
   # rmmod fiq_task
   rmmod: chdir(2.6.26): No such file or directory
   # mkdir /lib/modules/$(uname -r)
   # rmmod fiq_task
   fiq: got 4987481 FIQ events
   rmmod: module 'fiq_task' not found
   # grep '^fiq.task' /proc/modules
   # ls -l /lib/modules/$(uname -r)
   -rw-r--r--    1 root     root            0 Jan 11  2009 modules.dep.bb

With 1.9 I had no such problem, but I didn't investigate when the
problem was introduced. This is my current configuration, from "make
defconfig":

   CONFIG_DEFAULT_MODULES_DIR="/lib/modules"
   CONFIG_DEFAULT_DEPMOD_FILE="modules.dep"
   CONFIG_MODPROBE_SMALL=y
   CONFIG_FEATURE_MODPROBE_SMALL_OPTIONS_ON_CMDLINE=y
   CONFIG_FEATURE_MODPROBE_SMALL_CHECK_ALREADY_LOADED=y

Thanks
/alessandro