On 26/02/21 10:49AM, Arnd Bergmann wrote:

On Fri, Feb 26, 2021 at 9:16 AM 'Pratyush Yadav' via Clang Built Linux
[off-list ref] wrote:

quoted

Hi,

On 25/02/21 01:55PM, Arnd Bergmann wrote:

quoted

From: Arnd Bergmann <arnd@arndb.de>

Building this file with clang leads to a an unreachable code path
causing a warning from objtool:

drivers/spi/spi-rockchip.o: warning: objtool: rockchip_spi_transfer_one()+0x2e0: sibling call from callable instruction with modified stack frame

Use BUG() instead of unreachable() to avoid the undefined behavior
if it does happen.

Fixes: 65498c6ae241 ("spi: rockchip: support 4bit words")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/spi/spi-rockchip.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/spi/spi-rockchip.c b/drivers/spi/spi-rockchip.c
index 936ef54e0903..972beac1169a 100644
--- a/drivers/spi/spi-rockchip.c
+++ b/drivers/spi/spi-rockchip.c

@@ -521,7 +521,7 @@ static void rockchip_spi_config(struct rockchip_spi *rs,
               * ctlr->bits_per_word_mask, so this shouldn't
               * happen
               */
-             unreachable();
+             BUG();

checkpatch says:

  Avoid crashing the kernel - try using WARN_ON & recovery code rather
  than BUG() or BUG_ON()

Which makes sense to me. This is not something bad enough to justify
crashing the kernel.

I thought about rewriting it more thoroughly when I wrote the patch, but
couldn't come up with a good alternative, so I did the simplest change
in this direction, replacing the silent crash with a loud one.

Should we just dev_warn() and return instead, hoping that it won't do
more harm?

Hmm... I'm not very familiar with this device or driver so take what I 
say with a grain of salt. On the surface level it looks like it might 
end up doing something wrong or unexpected.

Returning an error code from this function (along with the dev_warn() or 
WARN_ON()) is the most sensible thing to do IMO. If the SPI layer sends 
an invalid value then the driver should be well within its rights to 
refuse the transaction. The function is currently void but making it 
return int seems fairly straightforward.

The backtrace from WARN_ON() probably doesn't help here.

       Arnd

-- 
Regards,
Pratyush Yadav
Texas Instruments Inc.

_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-rockchip