Re: [PATCH 0/14] init: deps: dependency based (parallelized) init

[Alexander Holler]

Am 17.10.2015 um 22:20 schrieb Greg Kroah-Hartman:
> On Sat, Oct 17, 2015 at 09:43:17PM +0200, Alexander Holler wrote:
> > Am 17.10.2015 um 20:38 schrieb Greg Kroah-Hartman:
> >
> > > So how long does that really take to call all probe functions in all
> > > possible order?  Real numbers please.  We have the tools to determine
> > > where at boot time delays are happening, please use them to find the
> > > problem drivers.
> >
> > No idea. You might ask Tomeu Vizoso (I've added him to cc) for details (or
> > search for the thread "On-demand device registration" where he complained
> > that his chromebook boots slow). I've just measured, that most my ARM boxes
> > booted faster when I've used the ordering, instead of slower through the
> > introduce overhead to order initcalls (without having parallelized the
> > initcalls).
>
> I've already asked him, I don't like his patch series to try to resolve
> this issue either :)
>
> > Posting times doesn't make much sense, as they heavily depend on the
> > configuration. Instead I've posted patches so you can test it yourself.
> >
> > But if you want a real time, my Netbook with a single core but HT Atom N270
> > boots in one second instead of two to "dmesg | grep Freeing".
>
> Try running the boot time graphic tool to determine where that time is
> spent, odds are you just need to enable a single driver to async it's
> probe function and you should be fine.
>
> Again, fix up the broken driver(s), don't paper over the issues with
> core changes that are not necessary.

I assume you are aware of the 80/20 rule. So you might see the 
parallelize feature of topological sort as an attempt to do some of the 
work in the last 20 percent left.

Sorry, but I've no idea why you are now trying to pin the stuff I'm 
talking about to a specific issue.

Or to talk in more clear words:

The current initcall ordering is, in my humble opinion, a whole and 
mostly undocumented mess. And I'm pretty sure it will become worse, I 
just have to wait and see. And if every attempt to fix that will be 
killed as fast as my one ...

Anyway, thanks for comments.

Regards,

Alexander Holler
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/