select() timeouts on large installations

[Artem Kachitchkine]

> I agree some kind of asynchronous probing could speed up detection
> in the general case - maybe even several devices in parallel?

In general, a parent device should be able to probe all its children 
devices in parallel.

 > (still trying to figure out what takes so much time)

Did you try to run hald with verbose logging on as David suggested. hald 
logger includes timestamps with millisecond granularity (I think) - 
should give some clues. To reduce the slowdown, redirecting stderr to 
ramdisk.

-Artem.