Re: [7/7,v8] NUMA Hotplug Emulator: Implement per-node add_memory debugfs interface

[Shaohui Zheng <shaohui.zheng@linux.intel.com>]

On Fri, Dec 10, 2010 at 03:30:38PM -0800, David Rientjes wrote:
> On Fri, 10 Dec 2010, Shaohui Zheng wrote:
> 
> > > That doesn't address the question.  My question is whether or not adding 
> > > memory to a memoryless node in this way transitions its state to 
> > > N_HIGH_MEMORY in the VM?
> > I guess that you are talking about memory hotplug on x86_32, memory hotplug is
> > NOT supported well for x86_32, and the function add_memory does not consider
> > this situlation.
> > 
> > For 64bit, N_HIGH_MEMORY == N_NORMAL_MEMORY, so we need not to do the transition.
> > 
> 
> One more time :)  Memoryless nodes do not have their bit set in 
> N_HIGH_MEMORY.  When memory is added to a memoryless node with this new 
> interface, does the bit get set?

When we use debugfs add_node interface to add a fake node, the node was created, 
and memory sections were created, but the state of the memory section is still 
__offline__, so the new added node is still memoryless node. the result of debugfs
add_memory interface doing the similar thing with add_node, it just add memory
to an exists node.

For the state transition to N_HIGH_MEMORY, it does not happen on the above too
interfaces. It happens when the memory was onlined with sysfs /sys/device/system/memory/memoryXX/online
interface.

That is the code path:
store_mem_state
	->memory_block_change_state
	 	->memory_block_action
			->online_pages

			if (onlined_pages) {
				kswapd_run(zone_to_nid(zone));
				node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
			}

does it address your question? thanks.

-- 
Thanks & Regards,
Shaohui

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>