Re: [PATCH: 002/017]Memory hotplug for new nodes v.4.(change name old add_memory() to arch_add_memory())

From: Yasunori Goto <y-goto_at_jp.fujitsu.com>
Date: 2006-03-22 17:06:52
> On Tue, 21 Mar 2006 17:08:18 -0800
> Dave Hansen <haveblue@us.ibm.com> wrote:
> > If I missed it before, please refresh my memory.  But, if we're
> > providing arch_nid_probe(addr), then why don't we just call it inside of
> > add_memory() on the start address, instead of in the generic code?
> > 
> I think just *probe* needs it. The firmware which supports memory-hotplug, ACPI, 
> i386/x86_64/ia64, can tell node number as pxm.(proximity domain)
> 
> add_memory() can pass address of paddr as args, but can't pass *pxm*. 
> 
> We already maintain pxm <-> nid map. But paddr <-> nid map isn't now.
> If add_memory() doesn't has nid as args, we have to maintain
> (1) pxm <-> nid map.
> (2) paddr <-> nid map.
> Becasue pfn_to_nid() is maintained by SPARSEMEM itself now, *new* paddr<->nid map
> is redundant, I think. And I think the firmware already has the map before calling
> add_memory(). 

In ACPI's case, kernel has to do 3 steps to get node id from physicall 
address.
  1) parse DSDT to get device handle of ACPI for new memory.
     This step can be described 3 detail steps.
      1-1) search memory device in DSDT.
      1-2) get _CRS of its memory device to see physicall address.
      1-3) compare 1-2)'s address with required paddr.
           (its memory device might be for other memory.)
           If it is for new memory, then goto step 2),
           else goto 1-1) again.
  2) get pxm against its device handle.
     It is just calling acpi_get_pxm().
  3) get node from pxm. If pxm is new one, new node id is assigned.

Step 1) is a bit complicated.
But, when notify of memory hot-add event reaches via ACPI,
its handle is already obtained. So, step 2) and 3) are enough to
get node id. If node id can be passed at add_memory(), that is all.
If not, kernel losts memory device handle information or node id once
at calling add_memory(), and search memory device handle again by step 1).



In addition, if paddr to node id translation is called in add_memory(),
then its code for -each arch- will be like followings.
This will be close from ugly "copy and paste" again....... :-(
 
add_memory(paddr)
{
	nid = paddr_to_node_id(paddr)
	if (!node_online(nid)) {
		pgdat = hotadd_new_pgdat(nid, start);
		if (!pgdat)
			return -ENOMEM;
	}
              :

Bye.

-- 
Yasunori Goto 


-
To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Received on Wed Mar 22 17:08:46 2006

This archive was generated by hypermail 2.1.8 : 2006-03-22 17:08:58 EST