> I'm trying to nail own a problem here, with my shiny new Maxtor 250Gb
> USB 2.0 disk. Under 2.4 (vanilla, latest 21-preX and 21-preX-acY) the
> disk will mount and talk nicely. As soon as any load hits it, e.g. a
> single cp from my internal CD-ROM to the disk, the mahcine load will
> sky-rocket and at some point within a few minuter hang the machine.
> 
> On 2.5 (vanilla and -mm) the load will show as i/o-wait and at some
> point hang any access to the drive, but the kernel will go on working.

That's a big clue -- nothing in the USB code ever shows up
as "i/o wait".  It can't, since USB is usually built as
modules and things like io_schedule() are, for some odd
reason, never exported for use in modules.  So USB I/O
can't use them, and won't show up as "i/o" ... and that
load must come from some place other than USB.

There are some patches that really ought to get merged
into the 2.4.21 release (2.5.70 has both of these), but
they shouldn't affect anything with an "i/o wait" symptom.

- Dave

[1] ehci-0429 ... applies to 2.4.21 with some fuzz, adds
     i/o watchdog timer, addressing both hardware (on some
     silicon) and software (suspected) irq lossage.

[2] ehci-0519 ... fixes some SMP-only problems.