Comment 405 for bug 88746

SNAFU

I've installed Fedora 10 Live CD to USB flash media.

Installed the i686 as well as the x86_64 version.

Installed on an nforce 430 platform and an SB600.

Installed to Kingston and Sandisk USB drives and to Sandisk SD card.

Every combination here is botched.

Decreasing max_sectors seems to alleviate the problem, but likely only because the probability of some timing error is decreased in that way.

Changing autosuspend and level parameters (in /sys/devices/pci.../*usb*/power/) also alleviates the problem, but doesn't resolve it.

When starting up, the Kingston (USB drive) media has it's power level set to "on". The EHCI host controller it's attached to however has it's power level set to "auto".

I've changed all power levels to "on" (also for the OHCI controllers). Although superfluous, I've set all 'autosuspend' values to -1. I've set the flash media's max_sectors to as low as 32, but have also extensively tried 128 and the default 240 (who came up with this value. Someone who watched the new time tunnel pilot?).

Booting the USB media, then running something like 'yum update' eventually generates an error akin to "unable to flush to dm-0" and then the dreaded "I/O Error, lost page write".

FWIW, "badblocks -w [drive]" reports no errors.

SB600 (laptop) is running on AC power. (nforce is a desktop).

Just the error (and a reboot) would mean a bug.

The fact that the entire system is deterministically non-bootable (I don't even want to attempt to salvage it) after these errors are reason for me to qualify this as a serious bug.

David