Red Hat Bugzilla – Bug 1967
lilo cannot read ramdisk size at boot time from boot device or kernel or initrd image
Last modified: 2008-05-01 11:37:49 EDT
Hi, I created an 8MB initrd manually, and placed it in /boot. I edited /etc/lilo.conf so it finds the kernel, initrd, etc. in /boot. I copied /etc/lilo.conf to /boot/etc/lilo.conf. I copied the hard disk devices needed for lilo from /dev/ to /boot/dev. I ran rdev -r <image> 8192, where <image> is the boot device, /dev/sda6, in my case. I ran it again with <image> being the kernel in /boot, and again with the initrd in /boot as <image>. I checked each with "rdev -r <image>", and they all reported back 8192 as expected. Then I run "lilo -c -r /boot". It appears to work fine. THE PROBLEM IS: unless I specify "ramdisk=8192" in /etc/lilo.conf, when I reboot, the kernel initializes 16 ramdisks of 4MB size, NOT ones that are 8MB size! This causes multiple I/O errors on dev 01:00, the ramdisk, as read requests go beyond the 4MB ramdisk boundary. I even set "root=current" in /etc/lilo.conf. (If I set "ramdisk=8192" in /etc/lilo.conf, rerun lilo, then reboot, I get 16 ramdisks of 8MB size, which is correct, but I shouldn't have to do this!) Background: I am writing a script to generate a more intelligent initrd, and I don't want to have to patch lilo.conf with a line "ramdisk=N", where N varies. The documentation for LILO says it is supposed to be able to read the ramdisk size from the boot image if no "ramdisk=N" line is given in /etc/lilo.conf, but this clearly doesn't work -- unless I am forgetting something! Please feel free to forward this email to the original LILO developer, or anyone else who may be helpful. thanks for your assistance!! Ed Ouellette techie@mit.edu
This is being forwarded to a developer for further review.
Here are more details: I may have found a bug in lilo, and/or I may have made a mistake in configuring an initial ramdisk. Here's the issue: lilo documentation says that if an initial ramdisk is enabled, and you DON'T include a line "ramdisk=N" in lilo.conf, that lilo will use the ramdisk size found in the boot image ( put there with rdev -r or compiled into the kernel, etc.). However, I have found that lilo ALWAYS defaults the ramdisk size to 4096K when this line is not present! Here's what may be happening (my guess): If lilo includes "ramdisk=0" in the kernel option line at bootup when lilo.conf does not contain "ramdisk=N", I can see how this happens, b/c then the kernel will use the default ramdisk size of 4096K. (I thought that if you ran "rdev -r <kernel> RAMDISK_SIZE" on the kernel, that it would "remember" its ramdisk size.... maybe the bug is in the kernel code?) I upgraded to version 21 of lilo, and lilo still has the same problem. When I include "ramdisk=N" in lilo.conf, everything works well, so I suspect maybe lilo is NOT checking (properly) that the bootimage has a non-default size before writing "ramdisk=0" to the kernel option line... or I may be misssing something major!
I don't think lilo is meant to do this.. The only documentation I can find is from user.tex: \item[\craw{ramdisk=}\meta{size}] Specifies the size of the optional RAM disk. A value of zero indicates that no RAM disk should be created. If this variable is omitted, the RAM disk size configured into the boot image is used. If I'm misunderstanding the situation, please reopen this bug and let me know what I'm missing.