Re: sparc64 bootup regression...

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Wed, Apr 30, 2008 at 04:15:40PM +0900, KAMEZAWA Hiroyuki wrote:
> On Tue, 29 Apr 2008 23:38:32 -0700
> Andrew Morton <akpm@linux-foundation.org> wrote:
> > On Tue, 29 Apr 2008 23:12:41 -0700 (PDT) David Miller <davem@davemloft.net> wrote:
> > > This commit causes bootup failures on sparc64:
> > > 
> > > commit 86f6dae1377523689bd8468fed2f2dd180fc0560
> > > Author: Yasunori Goto <y-goto@jp.fujitsu.com>
> > > Date:   Mon Apr 28 02:13:33 2008 -0700
> > > 
> > >     memory hotplug: allocate usemap on the section with pgdat
> > Thanks.  Does a straightforward revert fix it?  If so, we could do that while heads
> > are being scratched.

That patch broke the s390 port to SPARSEMEM (not upstream yet) as well.
Reverting it does help.
However increasing SECTION_SIZE_BITS to 25 (was: 20) does help too.
But that doesn't seem to help on sparc64 where it is already 30, strange.

> How about this ? If this is messy (or doesn't work), Goto-san will rework
> his own patch by himself. (this patch is against -mm but I think no HUNK to
> git tree)
> 
> ==
> This kind of page allocation, which specifies the address range,
> can fail easily.
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiruyo@jp.fujitsu.com>
> 
> Index: mm-2.6.25-mm1/mm/sparse.c
> ===================================================================
> --- mm-2.6.25-mm1.orig/mm/sparse.c
> +++ mm-2.6.25-mm1/mm/sparse.c
> @@ -264,10 +264,16 @@ static unsigned long *__init sparse_earl
>  	 * To solve above issue, this collects all usemap on the same section
>  	 * which has pgdat.
>  	 */
> +#ifdef CONFIG_NUMA /* contig_page_data for !NUMA case is not good to do this */
>  	section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
>  	usemap = alloc_bootmem_section(usemap_size(), section_nr);
>  	if (usemap)
>  		return usemap;
> +#endif
> +	/* above allocation can fail. */
> +	usemap = alloc_bootmem_node(NODE_DATA(nid), usemap_size());
> +	if (usemap)
> +		return usemap;
> 
>  	/* Stupid: suppress gcc warning for SPARSEMEM && !NUMA */
>  	nid = 0;

Fixes it on s390 at least.