On 06/14/2011 12:07 PM, Lauri Kasanen wrote: > Just saying I'm interested in this too. > > Our bootchart shows we spend a lot of time in modprobe [1], but since > the boot times were already "good enough", I haven't looked into it in > detail. My cursory googling showed that module loading in general is > slow, a fact people like Alan Jenkins and Carmelo Amoroso have tried to > change, but nothing is upstream yet. > > Is the goal to be compatible with util-linux binary files too?
This is probably a different issue you have. Generally modprobe gets accounted as using time for: a) deciding what to load (reading the alias files) b) actually loading the kernel module from disk; usually simple read() but might involve decompressing the modules c) invoking the system call so kernel starts using the module For Ed, the problem was that the system kernel is compiled with most modules built-in statically. This means that the system is generating lot of "modprobe" calls that actually end up doing nothing: most modules are built-in or not available. In this case the modprobe is doing just (a). While busybox modprobe takes only some milliseconds to do this, it piles up to lot after system has tried to execute modprobe 100 or more times. For you, based on the bootchart, it looks like modprobe is actually loading modules. and the time to do (a) is insignificant compared to (b) and (c). It's probably taking most of time in (c) which means kernel is working and initialising the module and possibly hardware. It's probably non-trivial to try to improve your situation. - Timo _______________________________________________ busybox mailing list busybox@busybox.net http://lists.busybox.net/mailman/listinfo/busybox
