Re: Can't load Nvidia driver on Fedora 9 x86_64
Hugh, I know it doesn't help much, but just to give you a positive, that driver is working on for my integrated GeForce 6150 on x86_64: $grep -i nvidia /var/log/Xorg.0.log (II) Module nvidia: vendor="NVIDIA Corporation" (II) NVIDIA dlloader X Driver 173.14.09 Wed Jun 4 23:48:23 PDT 2008 (II) NVIDIA Unified Driver for all Supported NVIDIA GPUs $lsmod | grep -i nvid nvidia 8108912 24 i2c_core 28448 2 nvidia,i2c_nforce2 $lspci -v | grep -i nvid 00:05.0 VGA compatible controller: nVidia Corporation C51PV [GeForce 6150] (rev a2) (prog-if 00 [VGA controller]) Kernel driver in use: nvidia Kernel modules: nvidiafb, nvidia $rpm -qa | grep nvidi xorg-x11-drv-nvidia-libs-173.14.09-1.lvn9.x86_64 kmod-nvidia-2.6.25.4-30.fc9.x86_64-173.14.05-3.lvn9.x86_64 (I have an old kernel still installed.) kmod-nvidia-173.14.09-1.lvn9.x86_64 kmod-nvidia-2.6.25.6-55.fc9.x86_64-173.14.09-1.lvn9.x86_64 xorg-x11-drv-nvidia-173.14.09-1.lvn9.x86_64 $rpm -qf /lib/modules/2.6.25.6-55.fc9.x86_64/extra/nvidia/nvidia.ko kmod-nvidia-2.6.25.6-55.fc9.x86_64-173.14.09-1.lvn9.x86_64 $uname -a Linux localhost.localdomain 2.6.25.6-55.fc9.x86_64 #1 SMP Tue Jun 10 16:05:21 EDT 2008 x86_64 x86_64 x86_64 GNU/Linux I guess it's a problem with the rebuilding of the module by the kmod-nvidia rpm, so I'd start by removing and re-adding it; and/or checking it's post/pre scripts and finding out how it rebuilds the module and doing so by hand just to get it working. I assume you have the akmods/kmodtool rpms installed? I believe they're related to livna's rebuilding process, but I haven't needed to confirm that. $ rpm -qa | grep kmod kmodtool-1-11.lvn9.noarch akmods-0.3.1-1.lvn9.noarch Regards, Richard -- fedora-list mailing list fedora-list@redhat.com To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
Re: Changing initrd contents and grub
On Mon, Jun 23, 2008 at 10:39:35AM -0600, Phil Meyer wrote: > Richard Michael wrote: >> Hello list, >> >> I've changed my RAID and LVM configuration and need to modify the >> respective commands in the /init of my initrd. >> >> I made those changes by decompressing and extracting the cpio archive, >> editting the init script (add a couple lines for mdadm, changed the >> activated volume group name), rebuilt a cpio archive (using the correct >> "-c"/"-H newc" SVR4 format) and fed it back through gzip (max >> compression), then I just moved aside the old initrd, replacing it with >> my new one: >> >> mkdir /boot/tmp >> cd !$ >> gzip -dc ../initrd- | cpio -id >> vi init >> find . -depth -print | cpio -oc | gzip -9 > ../initrd-.new cd .. >> mv initrd- initrd-.orig >> mv initrd-.new initrd- >> >> The kernel now panics (paraphrase) "can't find /init". >> >> It does not do this if I restore the original initrd. >> >> I have not changed the name of the initrd, filenames match grub.conf and >> grub's boot menu, etc. I have done this type of modification >> successfully in the past, but only changing a single character in /init. >> > > Just a thought here, since I have also tried this several times with > limited success: > > The whole point of mkinitrd is to avoid these 'by hand' operations. > > After you make your changes, run mkinitrd to generate a new initrd. It > will pick up changes in /etc/modprobe.conf and /etc/fstab and try to do the > right thing. Besides that, mkinitrd will accept arguments that allow > additional drivers to be loaded, with arguments if needed, as well as many > other options. I understand what you are saying, but I am of the opposite opinion. I have a decent picture of what needs to change, and where to change it. For example, those raid arrays which must be activated, and others not, etc. What I obviously need is a more detailed understanding of the kernel boot process, the BIOS and grub. (One problem here is troubleshooting the kernel and boot is tricky when the output moves so quickly on the screen!) Automated tools often frustrate me because (a) I don't learn anything, so I can't fix it when something goes wrong, and (b) something will go wrong due to all the guesswork of an automated tool. As I say, I've fixed problems with initrd's on other systems this way before. (In fact, I had to change the uuid of an array that was being activated. If I hadn't know anything about the contents of an initrd and how to modify it, etc. it would have been quite hard to fix with mkinitrd because the system wouldn't boot!) In this situation, mkinitrd is hard to employ because the system is booted from the Fedora DVD in rescue mode. I'm moving the entire system to a new configuration (RAID1 on RAID5). This means I can't run mkinitrd on *the* system to have it autoprobe, etc., etc. (chrooting from a rescue image always seems broken because the /dev entries never exist in the newly root tree). Moreover, I think nested RAID arrays will really confuse any of the automated tools because that configuration isn't (doesn't appear to be) supported (at least not at install time). Thanks though. Regards, Richard > I am pretty sure that a modern mkinitrd will make almost all need for > manual edits of an initrd image unnecessary. > > Good luck! > > -- > fedora-list mailing list > fedora-list@redhat.com > To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list -- fedora-list mailing list fedora-list@redhat.com To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
Changing initrd contents and grub
Hello list, I've changed my RAID and LVM configuration and need to modify the respective commands in the /init of my initrd. I made those changes by decompressing and extracting the cpio archive, editting the init script (add a couple lines for mdadm, changed the activated volume group name), rebuilt a cpio archive (using the correct "-c"/"-H newc" SVR4 format) and fed it back through gzip (max compression), then I just moved aside the old initrd, replacing it with my new one: mkdir /boot/tmp cd !$ gzip -dc ../initrd- | cpio -id vi init find . -depth -print | cpio -oc | gzip -9 > ../initrd-.new cd .. mv initrd- initrd-.orig mv initrd-.new initrd- The kernel now panics (paraphrase) "can't find /init". It does not do this if I restore the original initrd. I have not changed the name of the initrd, filenames match grub.conf and grub's boot menu, etc. I have done this type of modification successfully in the past, but only changing a single character in /init. So, it appears the kernel is not using my use initrd. Perhaps it is not prepared correctly (file magic for both new and old initrd files suggests they are the same, however)? Is grub involved somehow? Perhaps it can't find my new initrd? /boot is on a raid1 partition, ext2fs. Grub knows about this, and the system used to boot without problem. Any advice? Thanks, Richard -- fedora-list mailing list fedora-list@redhat.com To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
F9 install onto LVM on RAID1 on RAID5
Hello list, I am trying to install a new F9 system onto a RAID/LVM setup. As anaconda doesn't let me create the RAID/LVM configuration I require, I created these devices using mdadm and lvm in the shell during installation. I then created the filesystems and swap space, with labels, in the LVM volume group and I can mount them, read, write, etc.. So all is well with the underlying setup. However, returning to the install's "custom layout" partitioning page, anaconda displays the volume group and the names and sizes of the members are correct, but in the "TYPE" colume it indicates "foreign" and the mount point, and other fields are empty. If I click "LVM" or highlight one of the members and click "Edit" (to set the mount points and formatting options), anaconda responds with: -- Not enough physical volumes (...) Create a partition or RAID array of type "physical volume (LVM)" and click "LVM" again. -- Consequently, I cannot edit the member details to set mount points and formatting options and continue with the installation. How does anaconda determine the "type" of a RAID array; do md devices have types (as partitions do)? How can I satisfy it there are indeed physical volumes for an LVM (and that I have already configured them)? Alternately, how can I definitely tell anaconda to simply skip all partitioning and let me tell it which /dev entries to use for whichever partitions? Details === There are four disks in the system, I will add three more. The intention is the have a mirror three disks, for six total, plus one spare on one half. So: md0 is a raid1 (mirror) of four small partitions of each disk. mdadm --create /dev/md0 --level=raid1 --raid-devices=4 /dev/sd[abcd]1 md1 is a raid5 of the remaining portion of three disks plus a spare. mdadm --create /dev/md1 --level=raid5 --raid-devices=3 --spare-devices=1 --assume-clean /dev/sd[abcd]2 md4 is a raid1 (mirror), degraded because disks are currently missing, of md0 and "missing". mdadm --create /dev/md4 --level=raid1 --raid-devices=2 /dev/md0 missing md5 is a raid1 (mirror), degraded because disks are currently missing, of md1 and "missing". mdadm --create /dev/md5 --level=raid1 --raid-devices=2 /dev/md1 missing I configured my lvm volumes on md5, and as I mentioned, anaconda does see the members. lvm> pvcreate /dev/md5 lvm> vgcreate -s 32m vg0 /dev/md5 lvm> lvcreate -L 1024m -n root vg0 ; ... lvm> vgchange -a y vg0 Aside, anaconda displays md0 and md1 in the list of RAID volumes (both as type "foreign"), but *not* md4 and md5 -- even though they are just normal mirror RAIDs. Is this because they are degraded? I suspect anaconda lists the lvm members because it notices which vg's are active. It doesn't believe md5 contains a physical volume suitable for LVM use. (In fact, I don't think anaconda believes there any physical volumes for lvm on the system at all and, as above, it doesn't show md5 at all.) If I cannot get anaconda to cooperate, I'll install onto a raid5 array on temporary disks, then move the entire system into the proper nested-RAID5/RAID1/LVM setup. Thanks for suggestions. Regards, Richard -- fedora-list mailing list fedora-list@redhat.com To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list