2006-03-17 08:22:22

by Yasunori Goto

Subject: [PATCH: 003/017]Memory hotplug for new nodes v.4.(get node id at probe memory)

When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
before calling add_memory(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]>

arch/powerpc/mm/mem.c | 5 +++++
drivers/base/memory.c | 3 ++-
include/linux/memory_hotplug.h | 9 +++++++++
3 files changed, 16 insertions(+), 1 deletion(-)

Index: pgdat8/arch/powerpc/mm/mem.c
===================================================================
--- pgdat8.orig/arch/powerpc/mm/mem.c 2006-03-17 13:48:31.665710185 +0900
+++ pgdat8/arch/powerpc/mm/mem.c 2006-03-17 13:52:47.538753925 +0900
@@ -114,6 +114,11 @@ void online_page(struct page *page)
num_physpages++;
}

+int arch_nid_probe(u64 start)
+{
+ return hot_add_scn_to_nid(start);
+}
+
int __meminit arch_add_memory(int nid, u64 start, u64 size)
{
struct pglist_data *pgdata;
Index: pgdat8/include/linux/memory_hotplug.h
===================================================================
--- pgdat8.orig/include/linux/memory_hotplug.h 2006-03-17 13:48:31.626647685 +0900
+++ pgdat8/include/linux/memory_hotplug.h 2006-03-17 13:51:28.325864271 +0900
@@ -66,6 +66,15 @@ 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: pgdat8/drivers/base/memory.c
===================================================================
--- pgdat8.orig/drivers/base/memory.c 2006-03-17 13:47:31.558289046 +0900
+++ pgdat8/drivers/base/memory.c 2006-03-17 13:51:28.326840833 +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(arch_nid_probe(phys_addr),
+ phys_addr, PAGES_PER_SECTION << PAGE_SHIFT);

if (ret)
count = ret;

--
Yasunori Goto

2006-03-22 13:51:18

by Andi Kleen

Subject: Re: [PATCH: 003/017]Memory hotplug for new nodes v.4.(get node id at probe memory)

On Friday 17 March 2006 09:20, Yasunori Goto wrote:
> When CONFIG_NUMA && CONFIG_ARCH_MEMORY_PROBE, nid should be defined
> before calling add_memory(nid, start, size).

Seems a bit weird to have the node number assignment somewhere
hidden in memory hotadd. It would be probably cleaner to have
a separate step for this.

> +#if defined(CONFIG_NUMA) && defined(CONFIG_ARCH_MEMORY_PROBE)
> +extern int arch_nid_probe(u64 start);

Instead of adding such ugly ifdefs better just add stubs to all the
architectures that support memory hotplug. There are not that many
anyways.

-Andi

> +#else
> +static inline int arch_nid_probe(u64 start)
> +{
> + return 0;
> +}
> +#endif
> +