On Mon, 7 Oct 2002, David Brownell wrote: > It looks exactly like the oops I've seen sometimes when re-initializing > a host controller after rmmod ... usually ohci, since that's what I > was testing when I saw that. Curious that you see it even before > that first initialization, that changing init order makes things work > for you, and that it doesn't show up for you with other HCDs. Did you > happen to be loading this after the system was running multi-user, or > before? I'm reasonably sure I saw more problems when usbcore loaded > before multi-user.
I assume you are merely refering to the detailed init-script activities than to particular runlevel. I don't have the ehci-hcd included anywhere for automated loading. So the system came up with usbcore and ohci-hcd loaded. I've rmmod'ed the ohci-hcd and modprobed the ehci-hcd which was hitting the poisoned memory. Loading the ehci-hcd with the ohci still there was fine. To be 100% sure I've just done the following: - cold reboot into single user with nothing usb-related ever touched - modprobe usbcore - neither usbfs nor driverfs mounted at this point - modprobe ehci-hcd: works just fine without a problem - rmmod ehci-hcd - modprobe ehci-hcd again: result is exactly the same poisoned memory issue with the same backtrace that I've sent yesterday. Furthermore, I run into the same problem with any ohci-ehci combination if there is an intermediate rmmod. Looks like the real cause is in the module cleanup path and we are corrupting some list structure or similar. > I've seen a _lot_ of usbfs or driverfs problems in 2.5.40. So many > that I've stopped trying to use a CardBus adapter -- too many things So far, the ohci/ehci hcds are stable for me - but only until the first rmmod. > want to oops for me -- or unplugging/replugging USB devices more than > a couple times before I reboot. See this too. Situation improved for me if I do always keep the following order (read: workaround to avoid the symptoms): - wait until device is idle (module usecount==0) - unplug the device - rmmod the driver Any change to this scheme and I'm running into trouble sooner or later. Martin ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
