Re: drivers/pci/pcie/../pci.h:325:17: sparse: sparse: cast from restricted pci_channel_state_t

[Dan Carpenter]

On Wed, Mar 01, 2023 at 07:44:39AM +0300, Dan Carpenter wrote:
> > It seems they are only reported for architectures whose arch_xchg()
> > and arch_cmpxchg() macros cast the argument to an unsigned long.
> > Archictures such as x86 don't do that, but a number of others do.
> > The 0-day report, reproduced below in full, is for loongarch.
> > 
> > I'm wondering why the cast is necessary at all.  Digging in the
> > git history, I noticed that it has existed at least on arm since
> > forever.  I suspect that its use on newer arches such as loongarch
> > may be due to cargo-culting.
> > 
> 
> Speaking as an absolutely newbie and ignoramous, I can't see any point
> to the cast in arch_xchg().  But I am also surprised that silences the
> warning.  I would have thought that removing the cast would change the
> warning from "warning: cast from restricted my_type_t" to
> "warning: incorrect type in argument 1 (different base types)".

Your other option would be to add a __force to the cast.  I don't know
if this is a good option.  I guess first figure out if the __bit_wise
is really required.

regards,
dan carpenter

diff --git a/arch/arm64/include/asm/cmpxchg.h b/arch/arm64/include/asm/cmpxchg.h
index 497acf134d99..387c6af01941 100644
--- a/arch/arm64/include/asm/cmpxchg.h
+++ b/arch/arm64/include/asm/cmpxchg.h
@@ -93,7 +93,7 @@ __XCHG_GEN(_mb)
 ({									\
 	__typeof__(*(ptr)) __ret;					\
 	__ret = (__typeof__(*(ptr)))					\
-		__xchg##sfx((unsigned long)(x), (ptr), sizeof(*(ptr))); \
+		__xchg##sfx((__force unsigned long)(x), (ptr), sizeof(*(ptr))); \
 	__ret;								\
 })