Really? I wasn't aware of these problems. Perhaps this is the cause of my odd problems. I'm running Gentoo, and a day after upgrading to 2.6.12 (morph patchset), I had bugs. For one, the computer would hard lock if I unmounted my Windows partition certain ways. Since it's an entry in my /etc/fstab, I can choose to mount/unmount it via either absolute (mount /dev/hdb1 /mnt/windows) or relative (mount /mnt/windows). However, it would lock up if I used the relative method. Absolute worked fine. I blamed it on captiveNTFS and went on.
However, a couple of days later, all my modules refused to load. So no X (nvidia module, after all), no sound, and no network (note to self: always compile network into kernel. Troubleshooting is difficult without internet). The modules themselves are still there, and valid, but the kernel won't load them.
I'll try downgrading to 2.6.11 when I get home, and let you know.
I just checked the morph homepage, and there's a bugfix posted with the latest release... "(hopefully) fixed an inotify-related hang when umounting NTFS partitions". I might try this 'un before going back down to 2.6.11.
The latest gentoo ebuild for 2.6.12 has a lot of fixes for the kernel. The kernel you are using is either the vanilla package or is out of date.
The kernel is certainly stable - but maybe it's somewhat broken. If so, then just run two kernels until the latest patch release fixes all the problems. I just emerged the latest release a few days ago and I have had zero problems so far. Maybe you should consider using external modules for your drives, if they are available. I am pretty sure the kernel allows you to load those kinds of drivers as modules. I remember they used to make you run it built into the kernel image, but recently since not too many use the floppy or CD-ROM drives to boot the kernel, they have allowed the option of running all those device drivers as modules. This could indeed fix your problems, since often times these kinds of file system modules have conflict errors when run built in the kernel image.
PS: If I appear rambling in this post, it's most likely because I have had a tough day at work, which includes spilling coffee all over my groin on the subway and going uptown instead of downtown this morning. Ugh.
Just tested with morph6, and it works great. Both problems are gone. Now running happily again. I did have to re-emerge sys-fs/fuse (captiventfs dependency), although perhaps it's a requisite when you recompile your kernel. I can't remember. Anyway, other than that, no worries.
As for compiling drives as modules, I'll look into it. Personally, I've never noticed any performance differences between a modular and monolithic kernel.
Originally Posted by Welly Wu I am waiting for Red Hat engineers to release a less buggy 2.6.12-1.1xxx.FC4 kernel. The 2.6.12-1.1390.FC4 kernel is less buggy than 2.6.12-1.1.387.FC4 kernel, but both cause unexpected lockups compared to 2.6.11-1.1369.FC4 kernel.
I thought Fedora was following the path of Red Hat (Enterprise and the like). They use a custom made 2.4 kernel, where they've added in performance enhancing elements of 2.6, while staying with the stability of 2.4. Rather confusing when you don't know that and you're trying to upgrade it...