Hello Tony-san.
Fortunately, 2.6.15 includes memory hot-add function for i386 and ppc.
So, I made a patch for ia64.
This doesn't make new pgdat. All of new memory will belong to
node 0 by this patch. But this is simplest first step and best start for
future work.
I tested on my Tiger4. Please apply.
(This patch doesn't use ZONE_EASY_RECLAIM yet, because its zone will be useful
for just hot-remove.)
Signed-off-by: Yasunori Goto <[email protected]>
Index: new_feature_patch/arch/ia64/mm/init.c
===================================================================
--- new_feature_patch.orig/arch/ia64/mm/init.c 2006-01-05 15:43:10.000000000 +0900
+++ new_feature_patch/arch/ia64/mm/init.c 2006-01-05 20:23:00.000000000 +0900
@@ -635,3 +635,38 @@ mem_init (void)
ia32_mem_init();
#endif
}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+void online_page(struct page *page)
+{
+ ClearPageReserved(page);
+ set_page_count(page, 1);
+ __free_page(page);
+ totalram_pages++;
+ num_physpages++;
+}
+
+int add_memory(u64 start, u64 size)
+{
+ pg_data_t *pgdat;
+ struct zone *zone;
+ unsigned long start_pfn = start >> PAGE_SHIFT;
+ unsigned long nr_pages = size >> PAGE_SHIFT;
+ int ret;
+
+ pgdat = NODE_DATA(0);
+
+ zone = pgdat->node_zones + ZONE_NORMAL;
+ ret = __add_pages(zone, start_pfn, nr_pages);
+
+ if (ret)
+ printk("%s: Problem encountered in __add_pages() as ret=%d\n", __func__, ret);
+
+ return ret;
+}
+
+int remove_memory(u64 start, u64 size)
+{
+ return -EINVAL;
+}
+#endif
--
Yasunori Goto
On Fri, 2006-01-06 at 11:50 +0900, Yasunori Goto wrote:
> Fortunately, 2.6.15 includes memory hot-add function for i386 and ppc.
> So, I made a patch for ia64.
> This doesn't make new pgdat. All of new memory will belong to
> node 0 by this patch. But this is simplest first step and best start for
> future work.
It does look quite simple. Nice work.
> +#ifdef CONFIG_MEMORY_HOTPLUG
> +void online_page(struct page *page)
> +{
> + ClearPageReserved(page);
> + set_page_count(page, 1);
> + __free_page(page);
> + totalram_pages++;
> + num_physpages++;
> +}
You're the first one to get one of these in for an alternate
architecture. We'll need to keep an eye out so that one of these
doesn't pop up on each of the 64-bit arches with no highmem as we add
support. But, this should be just fine for now.
> +int add_memory(u64 start, u64 size)
> +{
> + pg_data_t *pgdat;
> + struct zone *zone;
> + unsigned long start_pfn = start >> PAGE_SHIFT;
> + unsigned long nr_pages = size >> PAGE_SHIFT;
> + int ret;
> +
> + pgdat = NODE_DATA(0);
> +
> + zone = pgdat->node_zones + ZONE_NORMAL;
> + ret = __add_pages(zone, start_pfn, nr_pages);
> +
> + if (ret)
> + printk("%s: Problem encountered in __add_pages() as ret=%d\n", __func__, ret);
For some reason, I thought we were officially supposed to use
__FUNCTION__ for stuff like this. However, I am usually lazy in my
debugging patches and use __func__. I'm a bad example.
This also looks a bit past 80 columns.
-- Dave
> On Fri, 2006-01-06 at 11:50 +0900, Yasunori Goto wrote:
> > Fortunately, 2.6.15 includes memory hot-add function for i386 and ppc.
> > So, I made a patch for ia64.
> > This doesn't make new pgdat. All of new memory will belong to
> > node 0 by this patch. But this is simplest first step and best start for
> > future work.
>
> It does look quite simple. Nice work.
>
> > +#ifdef CONFIG_MEMORY_HOTPLUG
> > +void online_page(struct page *page)
> > +{
> > + ClearPageReserved(page);
> > + set_page_count(page, 1);
> > + __free_page(page);
> > + totalram_pages++;
> > + num_physpages++;
> > +}
>
> You're the first one to get one of these in for an alternate
> architecture. We'll need to keep an eye out so that one of these
> doesn't pop up on each of the 64-bit arches with no highmem as we add
> support. But, this should be just fine for now.
>
> > +int add_memory(u64 start, u64 size)
> > +{
> > + pg_data_t *pgdat;
> > + struct zone *zone;
> > + unsigned long start_pfn = start >> PAGE_SHIFT;
> > + unsigned long nr_pages = size >> PAGE_SHIFT;
> > + int ret;
> > +
> > + pgdat = NODE_DATA(0);
> > +
> > + zone = pgdat->node_zones + ZONE_NORMAL;
> > + ret = __add_pages(zone, start_pfn, nr_pages);
> > +
> > + if (ret)
> > + printk("%s: Problem encountered in __add_pages() as ret=%d\n", __func__, ret);
>
> For some reason, I thought we were officially supposed to use
> __FUNCTION__ for stuff like this. However, I am usually lazy in my
> debugging patches and use __func__. I'm a bad example.
I didn't know it. Thanks!
> This also looks a bit past 80 columns.
Ok. This is newer one. :-)
Bye.
------------------
Fortunately, 2.6.15 includes memory hot-add function for i386 and ppc.
So, I made a patch for ia64.
This doesn't make new pgdat. All of new memory will belong to
node 0 by this patch. But this is simplest first step and best start for
future work.
I tested on my Tiger4. Please apply.
(This patch doesn't use ZONE_EASY_RECLAIM yet, because its zone will be useful
for just hot-remove.)
Signed-off-by: Yasunori Goto <[email protected]>
Index: new_feature_patch/arch/ia64/mm/init.c
===================================================================
--- new_feature_patch.orig/arch/ia64/mm/init.c 2006-01-07 10:58:27.000000000 +0900
+++ new_feature_patch/arch/ia64/mm/init.c 2006-01-07 10:59:52.000000000 +0900
@@ -635,3 +635,39 @@ mem_init (void)
ia32_mem_init();
#endif
}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+void online_page(struct page *page)
+{
+ ClearPageReserved(page);
+ set_page_count(page, 1);
+ __free_page(page);
+ totalram_pages++;
+ num_physpages++;
+}
+
+int add_memory(u64 start, u64 size)
+{
+ pg_data_t *pgdat;
+ struct zone *zone;
+ unsigned long start_pfn = start >> PAGE_SHIFT;
+ unsigned long nr_pages = size >> PAGE_SHIFT;
+ int ret;
+
+ pgdat = NODE_DATA(0);
+
+ zone = pgdat->node_zones + ZONE_NORMAL;
+ ret = __add_pages(zone, start_pfn, nr_pages);
+
+ if (ret)
+ printk("%s: Problem encountered in __add_pages() as ret=%d\n",
+ __FUNCTION__, ret);
+
+ return ret;
+}
+
+int remove_memory(u64 start, u64 size)
+{
+ return -EINVAL;
+}
+#endif
--
Yasunori Goto