We installed a new system from the RedHat Linux 5.2 CD. After the installation we installed the updated RPMS, including the 2.0.36-3 kernel. After making a new initrd image and changing /etc/lilo.conf we ran /sbin/lilo. It all went ok and we rebooted. After the reboot the system loaded the new kernel and the had a VFS panic and the system halted. Using a boot disk we could start the system and everything worked ok. Then we created a 2.0.36-3 bootdisk (mkbootdisk) and started from that. Another VFS panic... We booted from a rescue disc with the 2.0.36-3 bootdisk and examined the system. fdisk -l complained about partitions not ending on an even cylinder boundries. We backed up the system to a tape, repartitioned the system under 2.0.36-3 and restored the backup, ran a /sbin/lilo and the system worked alright. This system had an Adaptec 7985 onboard ultra scsi controller. Later we compiled a custom 2.0.36-3 kernel from the kernel-source package and removed a lot of features and included two of our "standard" scsi drivers in the kernel. We installed this kernel on four machines. Three with some Adaptec 2940 SCSI cards and one with some BusLogic card. The computers with the Adaptec cards had no problems but the computer with the BusLogic card got a lot of these errors when it used the new kernel: EXT2-fs warning (device 08:02): ext2_free_blocks: bit already cleared for block 4462 EXT2-fs warning (device 08:02): ext2_free_blocks: bit already cleared for block 4466 EXT2-fs warning (device 08:02): ext2_free_inode: bit already cleared for inode 186 scsi0: Adopting Geometry 64/32 from Partition Table and some files have been overwritten with other data. On this system fdisk -l doesn't compain about the partitions... I don't have the oppertunity to further investigate what might be wrong. Regards, Sebastian
Have you tried obtaining the 2.2 kernel source from from ftp://ftp.redhat.com/starbuck/i386/SRPMS/SRPMS/kernel-2.2.5-4.src.rpm. The aic7xxx support is better in the newer kernel than the one originally shipped with 5.2. Reopen this bug if this doesnt help.