|Summary:||PS/2 keyboard becomes dead/non-responsive on boot with kernel > 188.8.131.52-50.fc8|
|Component:||IEEE1394||Assignee:||Stefan Richter (stefanr)|
|Severity:||high||CC:||devzero, naveed, stefanr|
|Kernel Version:||> 184.108.40.206-50.fc8||Tree:||Fedora|
|Bug Depends on:|
|Attachments:||Diff -y on config-220.127.116.11 and config-18.104.22.168-92.fc8|
Description J 2008-05-25 12:36:01 UTC
Comment 1 J 2008-05-25 15:58:49 UTC
Building kernel 22.214.171.124 directly from the .kernel.org website using the make instructions from the readme and the config info inherited from 126.96.36.199.3-50 - not altering anything in the menuconfig stage, gave me a kernel which I could boot with no keyboard issues and make nvidia drivers for - a usable kernel. I will try kernel 188.8.131.52.7 from the website and if that works it must be down to a fedora related patch/mod.
Comment 2 J 2008-05-26 03:52:17 UTC
I have now built the 184.108.40.206 kernel and booted successfully. This would therefore point the problem at one of the fedora patches added to the 220.127.116.11 kernel. I'll now go through their patches and remove them from the rpm src until I can find the one at fault.
Comment 3 Roland Kletzing 2008-05-26 14:37:41 UTC
nice find ;) if this turns out to be a fedora issue - can't this be closed then ?
Comment 4 J 2008-05-27 02:48:25 UTC
I would suspect the answer is yes. Although I've found that you can't do make binrpm-pkg using the O option to specify an alternate path. If you do it fails to find the .config file. Instead you need to do make then make binrpm-pkg in the same src directory. It is also annoying that make install and make install_modules will create the *img file but the rpm doesn't. I don't suppose any of these comments really need to be in this bug? Before closing the bug would it be ok if I try and post the patch which is causing the fedora issue? There are 160 patches it may take me a while. Plus following the changelog comments makes it no easier a task.
Comment 5 J 2008-05-27 06:14:20 UTC
Before starting out on the patches I thought it would be worth comparing the 18.104.22.168 and 22.214.171.124-92 config files. (attached). Comparing this with the changelogs could activating I2C_PIIX4 be the culprit? Kernel.org Fedora # CONFIG_I2C_PIIX4 is not set | CONFIG_I2C_PIIX4=m I'll try the stock kernel with this turned on to see.
Comment 6 J 2008-05-27 06:16:19 UTC
Created attachment 16293 [details] Diff -y on config-126.96.36.199 and config-188.8.131.52-92.fc8 Diff command run on the two config files used to make the kernel. The non-fedora one works while the fedora one kills the keyboard at boot.
Comment 7 Naveed Hasan 2008-05-27 13:18:35 UTC
I am having a similar issue, however I cannot even Ctrl-Alt-Del as the PS/2 keyboard is completely unresponsive after "Enabling /etc/fstab swaps: OK" and can only hard reset the box. See https://bugzilla.redhat.com/show_bug.cgi?id=444694
Comment 8 J 2008-06-17 15:18:06 UTC
This may have been resolved. A patch has been put forward by the redhat/fedora guys it is suspected that the git firewire modules are to blame when coupled to a ALi firewire/usb combi card. adding blacklist firewire-ohci to /etc/modprobe.conf/blacklist allows the stock fedora kernel to boot. Likewise compiling the stock fedora kernel without the git firewire patches allows for a stable kernel. I think you can now close this.
Comment 9 Stefan Richter 2008-06-18 14:44:45 UTC
It is a regression in mainline too, post 2.6.25. Fix posted and verified with a newly acquired ALi based card: http://marc.info/?l=linux1394-devel&m=121380606431945 I will try to get it into mainline by the end of this week.
Comment 10 Stefan Richter 2008-06-18 14:46:45 UTC
Related: bug 10935 "fw-ohci: ALi M52xx unsupported"
Comment 11 Stefan Richter 2008-06-18 14:47:28 UTC
bug 10936 actually
Comment 12 Stefan Richter 2008-06-18 14:48:15 UTC
#-| no, 10935
Comment 13 Stefan Richter 2008-06-21 01:42:40 UTC
Fix merged into upstream 2.6.26-rc7. J and Naveed, thanks for locating the source of the bug. Fixing it was easy after you did all the debugging work.