Most recent kernel where this bug did not occur: 2.6.13 Distribution: Gentoo 2005.1 Hardware Environment: SMP Opteron 2x244, MSI K8T Master2 FAR,IBM PS/2 KB-8923 Software Environment: Linux Problem Description: I booted 2.6.14-rc1 to test if my external USB DVD would work better, so I burned a couple of DVDs with growisofs. I then rebooted, and the keyboard worked fine at the grub menu. Once I got to the 2.6.14-rc1 login prompt again, the keyboard was unresponsive. Reset and/or powering off the PC did not clear the problem- a poweroff followed by a 2.6.14-rc1 gets me to a login prompt with no keyboard response. To clear the problem, I had to reboot into 2.6.13 which gives me a responsive keyboard. I can then reboot into 2.6.14-rc1 and the keyboard responds at the login prompt. Steps to reproduce: 1) boot into 2.6.14-rc1 kernel 2) reboot into 2.6.14-rc1 kernel
Could you please try adding `usb-handoff' to the kernel boot command line?
Adding `usb-handoff' to the kernel boot command line fixes the problem. I was able to make the problem go away and subsequently reappear by adding and removing this option. Note I can only reproduce the problem if I do heavy USB I/O (burn a DVD via usb), and then reboot back into 2.6.14-rc1. I suspect one of the USB improvements in 2.6.14-rc1 that allow me to successfully burn a DVD has a slight regression. BTW, the bug I've submitted for the USB DVD burning is # 4373
There's a bugfix to EHCI reset/init processing that was in MM for a while and is now in the 2.6.15 queue, though I'd have preferred to see it in 2.6.14 instead ... that resolves a difference between the "usb-handoff" processing and normal EHCI reset/init code. (Whoever created the "usb-handoff" code did the wrong thing by not making the two instances of that logic be identical ...)
This should be fixed in 2.6.14, right?
I removed "usb-handoff" from my grub boot line, and haven't had any keyboard problems after a reboot after burning to my external USB burner. The problem appears to be fixed. However, looking at the 2.6.14 changelog it is unclear to me if the fix mentioned by David Brownell is in there or not. Perhaps my issue was fixed by some of the other EHCI changes.
Great, marking closed then.