Re: [PATCH] rsi: fix integer overflow warning

From: Joe Perches
Date: Thu Oct 05 2017 - 12:11:41 EST


On Thu, 2017-10-05 at 15:12 +0000, David Laight wrote:
> From: Joe Perches
> > Sent: 05 October 2017 13:19
> > On Thu, 2017-10-05 at 14:05 +0200, Arnd Bergmann wrote:
> > > gcc produces a harmless warning about a recently introduced
> > > signed integer overflow:
> > >
> > > drivers/net/wireless/rsi/rsi_91x_hal.c: In function 'rsi_prepare_mgmt_desc':
> > > include/uapi/linux/swab.h:13:15: error: integer overflow in expression [-Werror=overflow]
> > > (((__u16)(x) & (__u16)0x00ffU) << 8) | \
> > > ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
> > > include/uapi/linux/swab.h:104:2: note: in expansion of macro '___constant_swab16'
> > > ___constant_swab16(x) : \
> > > ^~~~~~~~~~~~~~~~~~
> > > include/uapi/linux/byteorder/big_endian.h:34:43: note: in expansion of macro '__swab16'
> > > #define __cpu_to_le16(x) ((__force __le16)__swab16((x)))
> >
> > []
> >
> > > The problem is that the 'mask' value is a signed integer that gets
> > > turned into a negative number when truncated to 16 bits. Making it
> > > an unsigned constant avoids this.
> >
> > I would expect there are more of these.
> >
> > Perhaps this define in include/uapi/linux/swab.h:
> >
> > #define __swab16(x) \
> > (__builtin_constant_p((__u16)(x)) ? \
> > ___constant_swab16(x) : \
> > __fswab16(x))
> >
> > should be
> >
> > #define __swab16(x) \
> > (__builtin_constant_p((__u16)(x)) ? \
> > ___constant_swab16((__u16)(x)) : \
> > __fswab16((__u16)(x)))
>
> You probably don't want the cast in the call to __fswab16() since
> that is likely to generate an explicit and with 0xffff.
> You will likely also get one if the argument is _u16 (not unsigned int).

It would just an explicit vs implicit cast as __fswab16 is
a static inline with a __u16 argument