Description of problem: /etc/init.d/pcmcia calls 'modprobe pcmcia_core.o' instead of 'modprobe pcmcia_core', which breaks. Version-Release number of selected component (if applicable): kernel-pcmcia-cs-3.1.31-13 How reproducible: Fully Steps to Reproduce: 1. install some pcmcia modules into /lib/modules/`uname -r`/pcmcia/ 2. reboot 3. Actual results: Bunch of "pcmcia: modprobe: Can't locate module" errors Expected results: successfully finds and loads the modules Additional info: This has been fixed in RHL/FC track, just needs merging into RHEL3. Here are the previous bugs: https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=81282 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=113796
Created attachment 104646 [details] patch to /etc/init.d/pcmcia for correct modprobe/insmod usage
In case any context is needed, this code path is only executed when /lib/modules/`uname -r`/pcmcia exists, which it doesn't for kernels installed via rpm. However, it does exist for kernels compiled from the kernel-source rpm. The other code path is fixed.
You say the other path is fixed, but can we not also get it fixed in case people do compile their own kernel? I know they would be unsupported, but they should have the freedom to do so.
The bug comes from Arjan's fix: * Tue Nov 26 2002 Arjan van de Ven <arjanv> - exclude range for Radeon ICP chipsets - use modprobe not insmod pcmcia-cs-3.1.18-script.patch: --- pcmcia-cs-3.1.31/etc/rc.pcmcia~ 2002-11-26 16:46:12.000000000 +0100 +++ pcmcia-cs-3.1.31/etc/rc.pcmcia 2002-11-26 16:46:12.000000000 +0100 @@ -104,9 +104,9 @@ KD=/lib/modules/`uname -r`/kernel/drivers/pcmcia if [ -d $PC ] ; then echo -n " modules" - /sbin/insmod $PC/pcmcia_core.o $CORE_OPTS - /sbin/insmod $PC/$PCIC.o $PCIC_OPTS - /sbin/insmod $PC/ds.o + /sbin/modprobe pcmcia_core.o $CORE_OPTS + /sbin/modprobe $PCIC.o $PCIC_OPTS + /sbin/modprobe ds.o elif [ -d $KD ] ; then /sbin/modprobe pcmcia_core /sbin/modprobe $PCIC Arjan, what's up with that? This is obviously bogus. Whatever that was supposed to fix, it never worked right. May I back it out?
backing out is the wrong thing, the reason the original code was wrong is that the modules changed location between 2.2 and 2.4 kernels, and this code assumed the 2.2 location (well technically it assumes the location pcmcia-cs puts it's OWN modules, while we want the kernel ones). right fix is to drop the .o
This appears to have been fixed in CVS in version RHEL-3/kernel-pcmcia-cs-3.1.31-14.src.rpm This is a small bug that will not hit default installs but would affect people who build their own modules or kernels. The customer ticket has been open a long time and they have verified that this would fix the problem. Deffinitly not a high priority, but obviously broken and already fiked in RHEL3 CVS