Re: ZONE_PADDING wastes 4 bytes of the new cacheline

From: Andrea Arcangeli
Date: Mon Oct 25 2004 - 19:19:13 EST


On Mon, Oct 25, 2004 at 10:49:32PM +1000, Nick Piggin wrote:
> The current stuff is pretty crufty. I think your changes are
> far better (aside from the minor fact they won't compile!).

;) yes, I still have to finish hooking the new code into the
page_alloc.c, this was just a preview.

dropping pages_xx would have been a bigger change so I'm avoiding it for
now since I agree I would need to add more fields than just min/low/high
for the realtime and GFP_HIGH features (and most important I couldn't
remove those 2 branches anywyas). In 2.4 I was ignoring __GFP_HIGH. I
doubt it makes much difference in practice, but it makes sense in theory
so I'll keep those new features of course (especially the RT one could
help in practice too).

> Also, switching to a calculation that has seen some real-world
> use would be a good idea before we think about turning it on
> by default.

agreed. the value tuning is quite simple to understand this way, it is
tuned exactly to start to reserve the entire lowmem on a 32G x86 box.
31G/32 = 0.9G > 800m.
-
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/