[PATCH] mtd: Fix kernel NULL pointer dereference in physmap.c

[H Hartley Sweeten]

On Tuesday, October 20, 2009 9:18 AM, Atsushi Nemoto wrote:
> On Tue, 20 Oct 2009 12:08:00 -0400, "H Hartley Sweeten" <hartleys at visionengravers.com> wrote:
>>>> Also, cleanup the #ifdef CONFIG_MTD_PARTITIONS stuff by using
>>>> mtd_has_partitions().
>>>
>>> And this cleanup cause build errors when CONFIG_MTD_PARTITIONS was not
>>> set.  A separate patch might be better for such cleanup.
>> 
>> Hmm..  Not sure why that would cause a build error.  Regardless, I will
>> remove that change from this patch.
>
> Thank you.  The build errors are something like:
>
> physmap.c:53: error: 'struct physmap_flash_info' has no member named 'nr_parts'
> physmap.c:174: error: 'part_probe_types' undeclared (first use in this function)

Ok.  That makes sense.

struct physmap_flash_info {
	struct mtd_info		*mtd[MAX_RESOURCES];
	struct mtd_info		*cmtd;
	struct map_info		map[MAX_RESOURCES];
#ifdef CONFIG_MTD_PARTITIONS
	int			nr_parts;
	struct mtd_partition	*parts;
#endif
};

#ifdef CONFIG_MTD_PARTITIONS
static const char *part_probe_types[] = { "cmdlinepart", "RedBoot", NULL };
#endif

I assumed that mtd_has_partitions() when CONFIG_MTD_PARTITIONS was not defined
would end up as something like this when compiled:

	if (mtd_has_partitions()) {
		/* ... */
	}

Would become:

	if (0) {
		/* ... */
	}

Then it would just optimze away.  It appears the code in the if (0) condition is
still parsed by the compiler so you get the errors above since the symbols are
#ifdef'ed out.

Oh well... I did post the updated patch based on your other comment.

Regards,
Hartley