mtd: fix random pointer dereference in OF device name handling

Here is the output from an mpc8548 based board.  There are
three instances of missing device name here:

1 |: Found 4 x16 devices at 0x0 in 32-bit bank. Manufacturer ID 0x008989 Chip ID 0x001818
2 |Intel/Sharp Extended Query Table at 0x0031
3 |Intel/Sharp Extended Query Table at 0x0031
4 |Using buffer write method
5 |cfi_cmdset_0001: Erase suspend on write enabled
6 |2 ofpart partitions found on MTD device
7 |Creating 2 MTD partitions on "":
8 |0x000000000000-0x000003f00000 : "space"
9 |0x000003f00000-0x000004000000 : "bootloader"

Lines 1 (BOL), 6 (EOL) and 7 (inside quotes) have the missing
device name issue.

Problem introduced with commit d68cbdd4fb04d2b756ad53c22f36943167b16340
"mtd: physmap_of: allow to specify the mtd name for retro compatiblity"

There are actually two bugs here.  The 1st is that mtd_name
is on the stack and never initialized.  It uses a call to
of_property_read_string() to get the pointer.  However this
function is explicitly documented as saying that the char
"...pointer is modified only if a valid string can be decoded."

Hence it isn't NULL, and we use a pointer off in the weeds as
the device name, leading to undefined behaviour.

The second issue is in the NULL check itself.  It uses a "?"
operator to choose between mtd_name and the devicetree based
name.  But the operator isn't given two choices.  One choice
(mtd_name) is missing from the RHS of the "?".

With these fixed, the output appears as follows:

1 |fc000000.flash: Found 4 x16 devices at 0x0 in 32-bit bank. Manufacturer ID 0x008989 Chip ID 0x001818
2 |Intel/Sharp Extended Query Table at 0x0031
3 |Intel/Sharp Extended Query Table at 0x0031
4 |Using buffer write method
5 |cfi_cmdset_0001: Erase suspend on write enabled
6 |2 ofpart partitions found on MTD device fc000000.flash
7 |Creating 2 MTD partitions on "fc000000.flash":
8 |0x000000000000-0x000003f00000 : "space"
9 |0x000003f00000-0x000004000000 : "bootloader"

All the names are now appearing where they should be.

Cc: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
---

[Introduced into mainline at v3.8-rc1~47 ; hence no need for stable Cc]

 drivers/mtd/maps/physmap_of.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

On Wed, 2013-01-23 at 13:59 -0500, Paul Gortmaker wrote:
> Here is the output from an mpc8548 based board.  There are
> three instances of missing device name here:

A similar fix is already in the linux-mtd.git tree.

On 13-02-04 03:05 AM, Artem Bityutskiy wrote:
> On Wed, 2013-01-23 at 13:59 -0500, Paul Gortmaker wrote:
>> Here is the output from an mpc8548 based board.  There are
>> three instances of missing device name here:
> 
> A similar fix is already in the linux-mtd.git tree.

OK, I see it in linux-next for 3.9 (commit 7dfe4be351e816) but
it needs to get to mainline/Linus before 3.8 gets released,
since that is where the regression is introduced (and that
is why I ended up not seeing the fix, and redoing it, as
I was working off of mainline).

$ git describe --contains d68cbdd4fb
v3.8-rc1~47^2~53

Thanks,
Paul.


>