Description of problem: Version-Release number of selected component (if applicable): kernel-2.6.12-1.1392_FC5 starting from kernel 2.6.12, while kernels 2.6.11 were working How reproducible: simply booting Actual results: mkrootdev: label /1 not found Mounting root filesystem mount: error 2 mounting ext3 Switching to new root ERROR opening /dev/console!!!!: 2 error dup2'ing fd of 0 to 0 error dup2'ing fd of 0 to 1 error dup2'ing fKernel panic - not syncing: Attempted to kill init! d of 0 to 2 unmounting old /proc unmounting ol Call Trace: d /sys switchroot: mo unt failed[<a000000100012160>] show_stack+0x80/0xa0 sp=e0000040ffd97c10 bsp=e0000040ffd90fb8 : 22 [<a000000100012af0>] dump_stack+0x30/0x60 sp=e0000040ffd97de0 bsp=e0000040ffd90fa0 Expected results: booting correctly Additional info:
Created attachment 115859 [details] Log of kernel boot
This is identical to my experience reported in bug #161347
So it seems an MPT Fusion adapter driver problem, and it's not related to IA64 architecture.
Try the defconfig or tiger_defconfig that were checked into Linus' GIT tree yesterday. They configure MPT Fusion correctly, and boot for me.
Also kernel-2.6.12-1.1395_FC5 is not working For Tony: try defconfig or tiger_defconfig means compile vanilla kernel using those files as .config ? How can I get last kernel sources with git ? Thank you.
Also kernel-2.6.12-1.1400_FC5.ia64 is not working
"try defconfig or tiger_defconfig" means: $ cp arch/ia64/defconfig .config $ make oldconfig $ make (or use arch/ia64/configs/tiger_defconfig). "How can I get last kernel sources with git?" ... if you want to delay learning git until it stabilizes, then you can just wait a day and any new code in Linus' GIT tree will be conveniently snap-shotted as a daily diff. See the "snapshot" link on the front page of www.kernel.org. If you are brave and want to try GIT ... which appears to be more than just a temporary thing ... then look for one of the recent "how-to" postings on LKML (e.g. search for "Updated git HOWTO for kernel hackers").
Vanilla 2.6.12.1 compiles and runs correctly with: CONFIG_FUSION=m CONFIG_FUSION_MAX_SGE=40 # CONFIG_FUSION_CTL is not set # CONFIG_FUSION_LAN is not set Anyway to have fusion disks as sda,sdb and sym53c8xx disks as sdc,sdd,sde,sdf I had to change modprobe.conf to this: alias scsi_hostadapter mptbase alias scsi_hostadapter1 mptscsih alias scsi_hostadapter2 sym53c8xx
*** Bug 161347 has been marked as a duplicate of this bug. ***
Note that rawhide is based on 2.6.13rc trees right now (The version number in the kernel filename ie kernel-2.6.12-1.1400_FC5 doesn't get bumped until /after/ the release of a new upstream revision). 2.6.12.1 may work, but changes may have occured in 2.6.13rc that have caused this driver to regress. You may want to test the latest upstream (currently 2.6.13rc2) and if that is similarly broken, report your findings to linux-scsi.org Thanks.
Same results for me running rawhide under vmware (the Fusion MPT driver is selected for vmware SCSI disks) on all kernels I've tried from 2.6.12-1.1434_FC5 up to and including 1441. Last working kernel I have for this VM is 2.6.11-1.1383_FC5.
This is still causing PANIC with kernel-smp-2.6.12-1.1485_FC5. Not good. Hope 2.6.13 final will have a fix.
SOLVED: The problem seems to be due to splitting of the mpt modules. I am not sure this is a bug or a feature. I have solved the problem by including "mptspi" module in the initrd image. /etc/modprobe.conf reads: alias scsi_hostadapter mptbase alias scsi_hostadapter1 mptscsih alias scsi_hostadapter1 mptspi install mptscsih /sbin/modprobe --ignore-install mptscsih install mptspi /sbin/modprobe --ignore-install mptspi with 2.6.12.x I did not need the two lines containing mptspi. If I do an lsmod I see that mptscsih module is now used by mptspi. I have a LSI U320 card with two processors. It seems like this should go into where module changes are handled under FC5.
The solution in comment #13 fixes the problem (still present with 2.6.12-1.1502_FC5) under vmware.
Fantastic. Yes, the solution in comment #13 also fixes the problem when booting on a Sun V20Z (dual Opteron, LSI 53c1030 Fusion-MPT SCSI). Basically, this bug affects all systems trying to boot from a SCSI disk requiring the MPT Fusion driver. It also fixed the boot with VMWare 5.5.0 RC1.
Just an extra note: This problem still exists with kernel 2.6.13-1.1589_FC5 The problem lies in the fact that the mptspi module is required, and yet not loaded automatically when doing 'modprobe mptscsih', which is why mkinitrd doesn't see the dependency and omits mptspi from the initrd file.
the fc5 installer/kudzu have been changed to create correct modprobe.conf's for the split up modules. FC5 should work out of the box. If this isn't the case with the latest rawhide builds (you can test installs with the boot.iso), please file an anaconda bug
I'm more concerned about people trying to upgrade from FC4 to FC5. I'll give that a shot and file an anaconda bug if necessary.
the installer should rewrite /etc/modprobe.conf on an upgrade, so this should be fine.
Anyone who has an up to date FC4 install has already gotten bitten by this problem (I did...) - 2.6.13-1.1526_FC4 needs modprobe.conf changing to reference mptspi.
FC4 has been dealt with in bug 169610