This is the ck-sources 1g_lowmem_i386 patch, changed accordingly to patch gentoo-dev-sources 2.10 kernel. Currently, gentoo-dev-sources can only use up to 896MB of RAM, but nowadays 1GB seems to be slightly common limit.
Steps to Reproduce:
1. Boot Linux
Warning only 896MB will be used.
Use a HIGHMEM enabled kernel.
896MB LOWMEM available.
Created attachment 51665 [details, diff]
Patch to use 1GB Low Memory
Use in /usr/src/linux.
This breaks things like vmware and cxoffice. If it is truly "the good solution" to the 1GB lowmem problem, then get it accepted into the official kernel tree first and then it will automatically be included in gentoo-dev-sources.
Can't speak to cxoffice, but the vmware problem is easy to fix. I have a diff for the vmware-config.pl that allows me to use vmware with the 1g_lowmem patch...
Should I attach?
Well, i agree that breaking applications isn't a good thing, but i wonder whether the problem is in the application itself or in the kernel change. I think the level of dependency between the kernel and the applications should be minimal, and so this kind of modification theoreticaly shouldn't be impeded by a set of applications.
I've submited this bug to kernel.org, it can be tracked in http://bugzilla.kernel.org/show_bug.cgi?id=4232.
Ted White, i think your patch would be useful, so anyone using the 1GB patch would know how to handle the vmware problem.
Created attachment 51724 [details, diff]
Patch to get VMWARE modules to compile when using 1g_lowmem
I have no problems running cxoffice 3.0 on a ck-sources kernel with 1gb-LOWMEM turned on.
I can confirm that cxoffice 4.1 works on a ck-sources kernel with 1gb-LOWMEM turned on.
Then get upstream (lkml.org) to accept it please and it will be in all further kernels.
*** Bug 92237 has been marked as a duplicate of this bug. ***
> This breaks things like vmware and cxoffice
It's these applications that are fundamentally broken, not the patch. I don't
see what the big issue is there because the patch provides a feature which is
entirely _optional_. Provided that the help text for the Kconfig option is clear
on the matter, then people are duly forwarned of the (very) limited scope for
breakage. Users of vmware/cxoffice can leave the feature disabled and use
HIGHMEM instead! In the meantime this is incredibly useful for people like me
who run servers with 1G of memory and don't wish to be encumbered with the
overhead of HIGHMEM. The patch works and is utterly stable. I have also tested a
variant patch I prepared which supports PaX (in fact that patch has also been
deployed by a developer for some time now) and it works perfectly. I suspect
that there would be resistance upstream on the basis of breaking pre-cooked
"binary" modules but this is a complete non-issue for gentoo. Would you not
re-consider this matter in view of its clear usefulness and proven stability?
No, because we generally just don't accept feature patches. If it is so
useful/correct/good, you should have no problem getting it into the mainline
kernel, right? :)
There is an open bug about this in kernel bugzilla.
So far nobody has responded though :(
You could try starting a discussion on the linux kernel mailing list.