From: Jeff Dike <jdike@ka...> - 2003-01-20 04:25:08
> Patch below makes SMP support in 2.5 UML functional again.
> It only works in tt mode, though (SKAS mode just immediately
> segfaults in get_current).
That's because can_do_skas was calling printk from a non-kernel stack.
> BTW, my Makefile trick to fool percpu checking logic might be
> called questionable, but I do not see clean and easy-to-implement way
> of resolving this issue ;)
I can think of worse things to call it :-)
The problem there is that UML builds vmlinux as a relocatable object, not
the final kernel. The awk script which checks that the percpu stuff is in
the right section operates on vmlinux, and expects the sections in it to be
correctly laid out. In UML, they weren't because that doesn't happen until
the linux link.
So, I just gave vmlinux enough of a linker script to put the kernel sections