?????: modprobe woes

[Timo Teräs]

Bernhard Fischer wrote:
> On Tue, Aug 05, 2008 at 09:18:45PM +0300, Timo Teräs wrote:
>> I can test performance of modprobe-small in real life tomorrow.
> 
> That would be helpful, yes. Perhaps you can provide profiles from
> module-init-tools vs. normal-modprobe vs. modprobe-new-from-vladimir ?

I can definitely give "time" results and also straces which give some
sort of idea where time is going. But for detailed stuff, I'm currently
profiling using callgrind. I haven't managed to get it working on my
slow box due to it running a hardened kernel. Needs some pax tweaking,
I suppose. If I get it working, I can post a detailed profile too.

>> I'd rather have the non-small (not too many hundred bytes larger) version
>> that is module-init-tools compatible and fast for my purposes.
>>
>> So I'm offering to write a module-init-tools compatible modprobe that
>> performs well and hopefully is smaller than the current "big" modprobe.
> 
> Well, very recently we started to grow an alternative module loader set
> in busybox (actually Vladimir did), and since this new one is not yet
> fully settled, it would be better to base off that one rather than
> spending time to tackle this problem for a third (or at least sixth, if
> you take modutils and module-init-tools into account) time.

Well, the Vladimir's code is not module-init-tools compatible, so
I do not currently consider that an option. It might be smaller, but for
me size is not the only factor.

I was hoping to replace the current "big" modprobe with something faster,
more compatible with module-init-tools and hopefully smaller too. So in
that sense it would be just an optimization of the current "big" one, but
it changes it quite a lot so it's not far from a new implementation either.

>> But can test it only on 2.6 kernel. Would it be accepted as a replacement
> 
> I personally do not need 2.0 nor 2.[12345] support. I'd say, just go
> for 2.6 and point people to the big utils that were recent when their
> kernel version was current and recommended. Of course that may be just
> me ;)

Would suit me. I can also try to do my best to make it support 2.4, as-in
do what the current code does, but I would not be able to test it.

- Timo