Built standard tarball 2.4.0 with ncr53c8xx scsi driver. There is IDE controller but no IDE drives (only SCSI). Built with modutils-2.4.0. During boot-up, gets to the point of looking for IDE, then message about unable to mount root_fs. Message shortly thereafter says it loaded the ncr53c8xx driver. Must assume that VFS code is not loading all potential disc controller modules before attempting initial mount, and will take a look versus 2.2 source. 1. Have done the mkinitrd/lilo steps 2. /etc/modules.conf: alias scsi_hostadapter ncr53c8xx alias eth0 old_tulip alias parport_lowlevel parport_pc alias usb-controller usb-uhci alias sound-slot-0 sb options sound dmabuf=1 options opl3 io=0x388 alias midi awe_wave post-install awe_wave /bin/sfxload /etc/midi/GU11-ROM.SF2 options sb io=0x240 irq=7 dma=0 dma16=6 mpu_io=0x300 3. Can send kernel build's .config file or /lib/modules/*/modules.dep on request. 4. Same behaviour, kgcc or gcc 2.96
You need to either compile the scsi-driver into the kernel, or include it in an initrd image and have lilo load it.
This isn't a bug in a Red Hat kernel, so I'm closing this ticket out. We don't support users compiling their own kernels because it would be too time consuming on our resources when you consider how complex a kernel build is to get completely right and how many people there are that have never done it before and would need hand holding to get through the process. Now, having said the appropriate disclaimer so I'm not sucked into a 4 page dissertation on the problem, the description you gave makes it sound like the ncr53c8xx module on the initrd is bad. Most likely, the initrd was loaded, you got the message from the linuxrc script on the initrd image stating that it was loading the ncr53c8xx module, but then never got any kernel messages from the ncr driver and then the machine failed to load the root fs and it hung. If that's a correct description, then your initrd image is bad. You can either compile the ncr driver directly into the kernel (the easy way to make sure things go right) or you can double back and redo the make modules; make modules_install steps on your kernel sources and then redo the mkinitrd step, likely making sure it is built by using the -f and -v flags to mkinitrd and visually observing that it does the right thing, then run lilo (verify all the file names in the lilo.conf file if needed) and then try again.