Re: [patch] x86: reduce srat verbosity in the kernel log

From: David Rientjes
Date: Fri Nov 13 2009 - 05:30:06 EST


On Fri, 13 Nov 2009, Ingo Molnar wrote:

> There's two problems outlined in this discussion:
>
> A) too verbose bootup that is annoying with 64 CPUs and a show-stopper
> with 4096 CPUs.
>
> B) the ad-hoc nature of our topology enumeration. Some of it is in
> /sys, some of it is in printk logs. None really works well and
> there's no structure in it.
>
> The simplest solution for (A) is what i suggested a few mails ago: dont
> print the information by default, but allow (for trouble-shooting)
> purposes for it to be printed when a boot option is passed.
>

Sigh, and even if that were done with a subsequent patch, you would still
want to reduce the debugging output from 1272 lines to 40, just like my
patch does without losing any information. It's insane to emit 1272
lines even if they are emitted only for a certain kernel parameter. I'm
sure we agree on that.

> Problem (B), topology info enumeration of a successful bootup is a
> different matter. It should be exposed to user-space via proper /sys
> abstractions, not via ad-hoc printks. There's ongoing work in that area,
> from Andreas Hermann, with patches posted. hpa expressed the view there
> that topology structure should be expressed via a nice vfs abstraction -
> i share that opinion.
>

Ingo, what do you want?

Your first criticism was that it should be limited only to a kernel
parameter but now it seems like you're insisting that the printk's get
removed completely and its exported via userspace. Then what is the
kernel parameter that you suggested for?

I'll leave the discussion with saying that if we still want to emit this
information with a parameter, that you'll still need to merge my patch at
some point to reduce the 1272 lines on Mark's system to 40. I'm unsure
why that isn't just merged right now since it's a clear improvement over
the current behavior, but I'm not going to beat a dead horse.

Thanks.
--
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/