[Patch:RFC] New zone ZONE_EASY_RECLAIM[0/5]

[Patch:RFC] New zone ZONE_EASY_RECLAIM[0/5]

[Yasunori Goto]

Hello.

I rewrote patches to create new zone as ZONE_EASY_RECLAIM.

Probably, many guys of here remember the discussion about
Mel-san's "fragmentation avoidance" patch in the last week.
In the discussion, another way was creating new zone like these patch
which I posted one years ago. 

http://sourceforge.net/mailarchive/forum.php?thread_id=5969508&forum_id=223

My motivation of creating new zone was for memory hotplug.
(Previous name of the zone was ZONE_REMOVABLE.)
It aimed to collect the page which was difficult for removing on few nodes,
and it made other nodes this zone to be removed easier.

But, I thought Mell-san's patch is also useful for this purpose.
So, I stopped my work about it after his patch was introduced.
Howerver unfortunately, Linus-san didn't agree his patches at last week.
So, I felt my old work should be restarted.

Frankly, I don't see that my patch set is/will be the best way
for others. My patch didn't concern about fragmentation.
And this patch has pros and cons as Mel-san said.
But I realize each persons are expecting against new zone for each reason.
So, I changed the name of new zone as ZONE_EASY_RECLAIM.
I wish this become good start point for them to be able to be happy.

This patch is for 2.6.14.

Please comment.

Thanks.

-- 
Yasunori Goto 


--
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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [Lhms-devel] [Patch:RFC] New zone ZONE_EASY_RECLAIM[0/5]

[Joel Schopp]

Yasunori Goto wrote:
> Hello.
> 
> I rewrote patches to create new zone as ZONE_EASY_RECLAIM.

Just to be clear.  These patches create the new zone, but they don't seem to 
actually use it to separate out removable memory, or to do memory remove.  I 
assume those patches will come later?  In any case this is a good start.

--
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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [Lhms-devel] [Patch:RFC] New zone ZONE_EASY_RECLAIM[0/5]

[Yasunori Goto]

> > I rewrote patches to create new zone as ZONE_EASY_RECLAIM.
> 
> Just to be clear.  These patches create the new zone, but they don't seem to 
> actually use it to separate out removable memory, or to do memory remove.  I 
> assume those patches will come later?  In any case this is a good start.

Yes.

Following patch is just to test the new zone on my ia64 box.
In this case, all nodes which doesn't have ZONE_DMA will be
ZONE_EASY_RECLAIM.
But, this should be more considered. At least, it should
check hotplug flag in SRAT table on ia64 (and x86-64?). 
Of course, the way of new zone has a issue of tuning among zones.
"Which area should be ZONE_EASY_RECLAIM" should be specified at boottime.
I'll try it next time.

Thanks.

Index: new_zone/arch/ia64/mm/discontig.c
===================================================================
--- new_zone.orig/arch/ia64/mm/discontig.c	2005-10-28 12:00:11.000000000 +0900
+++ new_zone/arch/ia64/mm/discontig.c	2005-11-07 20:10:25.000000000 +0900
@@ -663,9 +663,9 @@ void __init paging_init(void)
 
 		if (mem_data[node].min_pfn >= max_dma) {
 			/* All of this node's memory is above ZONE_DMA */
-			zones_size[ZONE_NORMAL] = mem_data[node].max_pfn -
+			zones_size[ZONE_EASY_RECLAIM] = mem_data[node].max_pfn -
 				mem_data[node].min_pfn;
-			zholes_size[ZONE_NORMAL] = mem_data[node].max_pfn -
+			zholes_size[ZONE_EASY_RECLAIM] = mem_data[node].max_pfn -
 				mem_data[node].min_pfn -
 				mem_data[node].num_physpages;
 		} else if (mem_data[node].max_pfn < max_dma) {


-- 
Yasunori Goto 


--
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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>