On Tue, 2009-10-06 at 18:01 +0200, Andi Kleen wrote:
> On Mon, Oct 05, 2009 at 11:19:24PM -0400, Lee Schermerhorn wrote:
> > [PATCH 11/11] hugetlb: offload [un]registration of sysfs attr to worker thread
> >
> > Against: 2.6.31-mmotm-090925-1435
> >
> > New in V6
> >
> > V7: + remove redundant check for memory{ful|less} node from
> > node_hugetlb_work(). Rely on [added] return from
> > hugetlb_register_node() to differentiate between transitions
> > to/from memoryless state.
> >
> > This patch offloads the registration and unregistration of per node
> > hstate sysfs attributes to a worker thread rather than attempt the
> > allocation/attachment or detachment/freeing of the attributes in
> > the context of the memory hotplug handler.
>
> Why this change? The hotplug handler should be allowed to sleep, shouldn't it?
Andy: perhaps it can. I'm not familiar with hotplug, so I followed a
pattern found elsewhere. I created a separate patch in case someone
familiar with this area says I don't need it.
>
> > N.B., Only tested build, boot, libhugetlbfs regression.
> > i.e., no memory hotplug testing.
>
> Yes, you have to because I know for a fact it's broken (outside your code) :)
We need to be able to remove all memory from a node without that node
disappearing [as I think it does on x86_64] to even exercise this code.
I think some ia64 platforms can do that, perhaps others.
Lee
--
To unsubscribe from this list: send the line "unsubscribe linux-numa" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
