On Thu, 2011-08-25 at 14:19 -0500, Christoph Lameter wrote:

On Thu, 25 Aug 2011, Peter Zijlstra wrote:

quoted

Also, I thought this_cpu thing's were at best locally atomic. If you
make them full blown atomic ops then even __this_cpu ops will have to be
full atomic ops, otherwise:


CPU0			CPU(1)

this_cpu_inc(&foo);	preempt_disable();
			__this_cpu_inc(&foo);
			preempt_enable();

might step on each other's toes.

They would both have their own instance of "foo". per cpu atomicity is
only one requirement of this_cpu_ops. The other is the ability to relocate
accesses relative to the current per cpu area.

Ah, but not if the this_cpu_inc() thing ends up being more than a single
instruction, then you have preemption/migration windows. Only when LL/SC
can deal with SC having a different EA from the LL and supports a big
enough offset could this possibly work.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>