Red Hat Bugzilla – Bug 50261
kudzu hangs with usb enabled
Last modified: 2007-04-18 12:35:16 EDT
Description of Problem:
Install Roswell with onboard usb disable in the bios. (diabled because
anaconda hangs if enabled, will report as separate problem).
After install, enable usb and booted the system. System hung hard during
kudzo processing and hard reset required to reboot.
If I bootup and Interactively NOT run kudzu, the system comes up with usb
enabled. I manually added the entry in /etc/modules.conf.
This is a TYAN Tiger 133 SMP motherboard with only ONE processor.
I will attach related files.
Steps to Reproduce:
1. see above
Created attachment 25313 [details]
Created attachment 25314 [details]
Created attachment 25315 [details]
Created attachment 25316 [details]
Created attachment 25317 [details]
Created attachment 25318 [details]
What happens if you do 'modprobe usb-uhci' from the command line?
*** Bug 50262 has been marked as a duplicate of this bug. ***
modprobe usb-usci works .. see attached lsmod output.
Created attachment 25351 [details]
lsmod before modprobe
Created attachment 25352 [details]
lsmod after modprobe
Note: no usb devices are attached to this system.
If, after boot, you run 'strace kudzu', where does it hang?
This defect is considered SHOULD-FIX for Fairfax
Because of some hardware changes I made, I am currently unable to reproduce this
problem ... usb now initializes fine!
I swapped the motherboards between two systems and the problem went away.
The motherboard/configuration that was having the problem (TYAN) had an AGP
video adapter and a netgear NIC. This same motherboard with a PCI video adapter
and a 3COM 905B NIC works fine now.
It will be a major headache to switch things back so I would prefer to drop this
but will give it a try if you folks want me to.
Over to you.
kudzu doesn't do anything that would crash the machine except load the
usb controller module. If that hangs the machine, it's a kernel bug.
So, assigning this to the kernel. They can decide whether they want yo
to try and reproduce it or not.
The problem is that when the hang was occuring that I could manually modprobe
usb-uhci and it would work fine ... it only hung (hard reset needed) in kudzu or
in anaconda (which I assume used some kudzu code).
Kudzu kills machines often because it removes modules.
This area was double problematic with USB because
USB has a separate thread (khubd) which is triggered
when HC driver loads. Then kudzu unloads module immediately
and khubd blows up.
I think I plugged all holes in hub.c for 7.1, but perhaps
something got regressed in new -ac kernels.
It may be possible to reproduce by running something
insmod usb-uhci && rmmod usb-uhci # trips khubd
If anyone can reproduce this bug (or 50262), reopen!
I need to see it BEFORE WE FREEZE 7.2.