x86_64/kdump crashkernel reservation breakage in 2.6.31-rc

x86_64/kdump crashkernel reservation breakage in 2.6.31-rc

[Chris Wright]

I hesitate to label this a regression since 2.6.30, but at least user
visible change since 2.6.30 that at a minimum needs some docs updates.

The commit "ceefccc: x86: default CONFIG_PHYSICAL_START and
CONFIG_PHYSICAL_ALIGN to 16 MB" has a side effect for kdump users.
The problem is a little convoluted because the typical setup for kdump
includes:

crashkernel=64M@16M or crashkernel=128M@16M

These uses don't account for a relocatable kernel which could omit the
@X location specifier, but it's the typical documented method[1][2][3].

Trouble is, now the original kernel loads at 16MB since it's alignment
request will push it up from the old default of 2MB, and the existing
attempt to reserve a region for the crashkernel will fail.

I think the docs are the problem here, in that they don't adequately
prepare a user for the (now common) relocatable kernel case.
Alternatively, back out the CONFIG_PHYSICAL_ALIGN change...

Thoughts?

thanks,
-chris

[1] http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/kdump/kdump.txt;hb=HEAD#l271
[2] http://fedoraproject.org/wiki/Kernel/kdump
[3] http://kbase.redhat.com/faq/docs/DOC-6039