Re: please revert bogus patch to vmscan.c

From: Benjamin LaHaise (bcrl@redhat.com)
Date: Mon Oct 29 2001 - 21:25:46 EST


On Mon, Oct 29, 2001 at 05:42:07PM -0800, Linus Torvalds wrote:
>
> On Mon, 29 Oct 2001, David S. Miller wrote:
> >
> > I'm asking him to show the case that "breaks for something
> > else".
>
> Guys, guys, calm down.

I'm trying to be. Dave just rubs me the wrong way with his inability to
think about the problem for a minute instead of insisting on proof that
incorrect behaviour can happen. Damnit, it's supposed to be a kernel on
its way towards being more stable, not less. This is especially insulting
when you can bother to go through the work of it as a thought experiment in
less than 5 minutes to realise what could happen.

Dave: please read the above paragraph again. Now, can you see why I'm
arguing for doing the optimization in the *correct* way first? The
microbenchmark will always "prove" that avoiding the tlb flush is a win.
The test to prove that the failure case can happen is non-trivial, and
proving the real world win/loss is lengthy task involving lots of
benchmarks and effort. Please just sit down and think for 5 minutes and
acknowledge that it is a possibility!

> The difference in call frequency would, on large machines, probably be on
> the order of several magnitudes, which will certainly cut the overhead
> down to the noise while satisfying people who have architectures that can
> cache things for a long time.
>
> Agreed?

Which is exactly what was in the back of my mind in the first place. I
didn't write the patch as the distraction of going off on yet another
tangent when I'm this > < close to being done with the battle I'm currently
in just doesn't make sense. I'm sorry, but I'm not the best person for
doing a brain dump when in the middle of something, plus I assume that
people can think about the how and why of things for themselves. You'll
note that I never denied that the microoptimization in question is a win;
I fully well expect it to be. However, from the point of view of stability
we *want* to be conservative and correct. If Al had to demonstrate with
'sploits that a possible vfs race could occur every time he found one,
wouldn't he be wasting time that could be better spent finding and fixing
other problems??? Dave, please accept that other people's opinions
occasionally hold value and reconsider reacting negatively without thinking
first.

                -ben

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



This archive was generated by hypermail 2b29 : Wed Oct 31 2001 - 21:00:38 EST