[plug] initrd confusion!

Denis Brown dsbrown at cyllene.uwa.edu.au
Fri Apr 16 11:28:12 WST 2004

Dear PLUG list members,

I have a situation where I need to have a modularised kernel, specifically 
with the RAID device as a module.   The reason for this, is to allow the 
low level drive monitoring software to work.   I have been reading up on 
initrd but find that I am confused, as follows...

0. There are no IDE or ATA devices in this box, only SCSI in hardware RAID.
1. The RAID driver will be compiled into the kernel as a module.   So the 
kernel will not be able to see any file systems until the module is loaded.
2. The module will be inaccessible to the kernel because, well, it cannot 
be loaded!
3. Initrd would seem to offer help.   Initrd basically, if I understand it 
correctly, will load a kernel into ramdisk, load the required RAID driver 
module and then switch over to the "real" file system.

BUT... if initrd can see the file system in order to load the RAID driver 
module, how is initrd's "kernel" different from the normal kernel?   I will 
be compiling the kernel with both initrd+ramdisk and modularised RAID 
driver support.   It seems to me that the kernel will somehow be 
schizophrenic - sees the RAID via a module but at boot time can see the 
RAID anyway.   Huh???

Earlier this year I began to set up a couple of servers with IBM/Advantec 
ServeRAID controllers.   The RAID drivers were compiled into the 
kernel.   The systems run fine but I would like to monitor the health of 
the physical drives via the ipssend utility.   Ryan helped (around March) 
by pointing me to alternative ipssend versions but I continued to draw 
blank looks from ipssend, which claimed that no controllers could be 
found.   I finally had time to look through the dumplog scripts and 
determined that ipssend looks for the ips.o module, which in my present 
configuration is absent.

So, either I need to come to grips with initrd - which is possible except 
for the mental burble above - or find another way to communicate with the 
RAID mechanism to examine individual drive statuses (stati?)   I am 
perfectly happy - in fact happier - to have the RAID driver compiled in to 
the kernel since it is unlikely to change (be updated) frequently and I'd 
hate to have a situation where it somehow automagically got rmmod'ed :-).

TIA for any insights,

More information about the plug mailing list