Re: PS3: trouble with SPARSEMEM_VMEMMAP and kexec

[Geoff Levand <geoffrey.levand@am.sony.com>]

Andrew Morton wrote:
> On Wed, 5 Dec 2007 10:52:48 +0100 (CET)
> Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com> wrote:
> 
>> --------------------------------------------------------------------------------
>> Subject: sparsemem: sparse_add_one_section() may fail to allocate memory
>> 
>> sparsemem: sparse_add_one_section() may fail to allocate memory, and must check
>> whether the allocation succeeded before proceeding to touch the allocated
>> memory.
>> 
>> From: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
>> 
>> Signed-off-by: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
>> ---
>> FIXME There are still some possible memory leaks in sparse_add_one_section():
>>   - usemap is never deallocated
>>   - __kfree_section_memmap() is a not yet implemented dummy
> 
> I already had
> 
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.24-rc4/2.6.24-rc4-mm1/broken-out/mm-sparsec-improve-the-error-handling-for-sparse_add_one_section.patch


This one has an error in it.  A patch to fix it is below.


> and
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.24-rc4/2.6.24-rc4-mm1/broken-out/mm-sparsec-check-the-return-value-of-sparse_index_alloc.patch
> 
> queued.  Do they fix the problem, and should they be merged in 2.6.24?


These two plus my fix below allow the hot plug add_memory() call to fail
gracefully and for the platform code to continue to boot on the
128MB of boot mem.

With ps3_defconfig the condition is only hit by the second stage
kexec'ed (kboot) kernel, which is not generally built by end users,
but there is a chance this condition would be hit by custom kernel
config, so I think they should go in for 2.6.24.

I'll continue to work on a fix for the memory allocation failure.

-Geoff


------------------
Subject: sparsemem: Fix sparse_index_init return check

sparse_index_init() returns -EEXIST to indicate the index
has already been created.  Exclude this from the error check
on the return value.

Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
---
 mm/sparse.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -392,7 +392,7 @@ int sparse_add_one_section(struct zone *
 	 * plus, it does a kmalloc
 	 */
 	ret = sparse_index_init(section_nr, pgdat->node_id);
-	if (ret < 0)
+	if (ret < 0 && ret != -EEXIST)
 		return ret;
 	memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
 	if (!memmap)