Hi Russell,

	Patch against 2.6-test5.

	Kills a unneeded set_current_state (TASK_RUNNING) after 
schedule_timeout and replaces a TASK_RUNNING settings with 
non-memory-barrier versions.

	On uart_suspend_port, that set state thing isn't needed if 
schedule_timeout was executed..but I couldn't come up with a (clean) 
fix for that, since we need to execute ops->shutdown. So I just 
replace it with __set_current_state so It's just an attribution and 
it's not that expensive.

	Feel free to tell me I screwed up :)

	Thanks,

Felipe