When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
before calling add_memory_node(nid, start, size).
Each arch , which supports CONFIG_NUMA && ARCH_MEMORY_PROBE, should
define arch_nid_probe(paddr);
Powerpc has nice function. X86_64 has not.....
Note:
If memory is hot-plugged by firmware, there is another *good* information
like pxm.
Signed-off-by: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Yasunori Goto <[email protected]>
Index: pgdat6/arch/powerpc/mm/mem.c
===================================================================
--- pgdat6.orig/arch/powerpc/mm/mem.c 2006-03-06 19:34:53.000000000 +0900
+++ pgdat6/arch/powerpc/mm/mem.c 2006-03-06 19:39:51.000000000 +0900
@@ -114,6 +114,11 @@ void online_page(struct page *page)
num_physpages++;
}
+int __devinit arch_nid_probe(u64 start)
+{
+ return hot_add_scn_to_nid(start);
+}
+
int __devinit arch_add_memory(int nid, u64 start, u64 size)
{
struct pglist_data *pgdata;
Index: pgdat6/include/linux/memory_hotplug.h
===================================================================
--- pgdat6.orig/include/linux/memory_hotplug.h 2006-03-06 19:34:47.000000000 +0900
+++ pgdat6/include/linux/memory_hotplug.h 2006-03-06 19:40:57.000000000 +0900
@@ -63,6 +63,15 @@ extern int online_pages(unsigned long, u
/* reasonably generic interface to expand the physical pages in a zone */
extern int __add_pages(struct zone *zone, unsigned long start_pfn,
unsigned long nr_pages);
+#if defined(CONFIG_NUMA) && defined(CONFIG_ARCH_MEMORY_PROBE)
+extern int arch_nid_probe(u64 start);
+#else
+static inline int arch_nid_probe(u64 start)
+{
+ return 0;
+}
+#endif
+
#else /* ! CONFIG_MEMORY_HOTPLUG */
/*
* Stub functions for when hotplug is off
Index: pgdat6/drivers/base/memory.c
===================================================================
--- pgdat6.orig/drivers/base/memory.c 2006-03-06 19:16:37.000000000 +0900
+++ pgdat6/drivers/base/memory.c 2006-03-06 19:39:51.000000000 +0900
@@ -310,7 +310,8 @@ memory_probe_store(struct class *class,
phys_addr = simple_strtoull(buf, NULL, 0);
- ret = add_memory(phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
+ ret = add_memory_node(arch_nid_probe(phys_addr),
+ phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);
if (ret)
count = ret;
Index: pgdat6/arch/x86_64/mm/init.c
===================================================================
--- pgdat6.orig/arch/x86_64/mm/init.c 2006-03-06 19:34:53.000000000 +0900
+++ pgdat6/arch/x86_64/mm/init.c 2006-03-06 19:39:51.000000000 +0900
@@ -493,6 +493,11 @@ void online_page(struct page *page)
num_physpages++;
}
+int arch_nid_probe(u64 start)
+{
+ return 0;
+}
+
int arch_add_memory(int nid, u64 start, u64 size)
{
struct pglist_data *pgdat = NODE_DATA(nid);
--
Yasunori Goto
Yasunori Goto <[email protected]> wrote:
>
> When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
> before calling add_memory_node(nid, start, size).
>
> Each arch , which supports CONFIG_NUMA && ARCH_MEMORY_PROBE, should
> define arch_nid_probe(paddr);
>
> Powerpc has nice function. X86_64 has not.....
This patch uses an odd mixture of __devinit and <nothing-at-all> in
arch/x86_64/mm/init.c. I guess it should be using __meminit
throughout.
> Yasunori Goto <[email protected]> wrote:
> >
> > When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
> > before calling add_memory_node(nid, start, size).
> >
> > Each arch , which supports CONFIG_NUMA && ARCH_MEMORY_PROBE, should
> > define arch_nid_probe(paddr);
> >
> > Powerpc has nice function. X86_64 has not.....
>
> This patch uses an odd mixture of __devinit and <nothing-at-all> in
> arch/x86_64/mm/init.c. I guess it should be using __meminit
> throughout.
Oh... I made mistake. I'll fix them.
--
Yasunori Goto
> > Yasunori Goto <[email protected]> wrote:
> > >
> > > When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
> > > before calling add_memory_node(nid, start, size).
> > >
> > > Each arch , which supports CONFIG_NUMA && ARCH_MEMORY_PROBE, should
> > > define arch_nid_probe(paddr);
> > >
> > > Powerpc has nice function. X86_64 has not.....
> >
> > This patch uses an odd mixture of __devinit and <nothing-at-all> in
> > arch/x86_64/mm/init.c. I guess it should be using __meminit
> > throughout.
>
> Oh... I made mistake. I'll fix them.
Hmmm. I'm confusing again about this. :-(
Dave-san, Joel-san.
Why does Powerpc use __devinit for add_memory()?
Usually, add_memory() is never called at boottime.
So, I suppose __meminit nor __devinit is not needed at all around here.
But, does it have a plan that add_memory() is called only boottime on
Powerpc?
--
Yasunori Goto