Bug#295678: kernel-image packages with CONFIG_HIGHMEM64G=y enabled
On Thu, 2006-02-16 at 18:12 +0100, Yann Rouillard wrote:
> > I can't say I'm a fan of adding another flavour. How many people are
> > actually using >4GB of memory on x86? I suspect (or hope) that people
> > who are going to be doing those sorts of things will be using x86_64
> > and ia64 hardware. Christian's problem, aiui, is simply that the kernel
> > sees 3.5GB instead of 4GB. I'm not sure whether this is by design, or is
> > a bug, but I'd much rather see that fixed instead.
> This is not really a bug, see:
> The 3.5GB-4GB address space is reserved for PCI devices, so, as only 4GB
> can be adressed on 32bits proc, there is an address space shortage which
> prevents using the 3.5GB-4GB RAM.
> So the only solution is to enable CONFIG_HIGHMEM64G to increase the
> address space so that the 3.5GB-4GB RAM become addresseable.
> However I would be interested in some numbers/benchmarks on the
> performance penalty of the PAE mode. Is this worth 512MB of additionnal
> RAM ?
There was certainly a performance penalty last I checked (~1 year ago),
but I don't have the numbers handy to back that up. I'd personally be
happy with an additional pae flavor, but not changing the default.