On Thu, May 26, 2016 at 01:18:21AM +0200, Daniel Schwierzeck wrote:
Am 17.05.2016 um 12:56 schrieb Paul Burton:
When building for MIPS64 and providing a pointer to _ACAST32_, optionally via CPHYSADDR or one of the CKSEGxADDR macros, the cast directly to a 32 bit int leads to compilation warnings such as the following:
In file included from ./arch/mips/include/asm/io.h:17:0, from drivers/net/pcnet.c:14: drivers/net/pcnet.c: In function ‘pcnet_virt_to_mem’: ./arch/mips/include/asm/addrspace.h:39:29: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast] #define _ACAST32_ (_ATYPE_)(_ATYPE32_) /* widen if necessary */ ^ ./arch/mips/include/asm/addrspace.h:51:25: note: in expansion of macro ‘_ACAST32_’ #define CPHYSADDR(a) ((_ACAST32_(a)) & 0x1fffffff) ^ ./arch/mips/include/asm/addrspace.h:71:25: note: in expansion of macro ‘CPHYSADDR’ #define CKSEG0ADDR(a) (CPHYSADDR(a) | CKSEG0) ^ drivers/net/pcnet.c:144:23: note: in expansion of macro ‘CKSEG0ADDR’ virt_addr = (void *)CKSEG0ADDR(addr); ^
Fix this by first casting provided values to a pointer-width integer, then truncating to a 32 bit int & widening back to pointer-width.
Signed-off-by: Paul Burton paul.burton@imgtec.com
arch/mips/include/asm/addrspace.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/mips/include/asm/addrspace.h b/arch/mips/include/asm/addrspace.h index 0994e96..fe497b5 100644 --- a/arch/mips/include/asm/addrspace.h +++ b/arch/mips/include/asm/addrspace.h @@ -36,8 +36,8 @@ #define _ACAST32_ #define _ACAST64_ #else -#define _ACAST32_ (_ATYPE_)(_ATYPE32_) /* widen if necessary */ -#define _ACAST64_ (_ATYPE64_) /* do _not_ narrow */ +#define _ACAST32_ (_ATYPE_)(_ATYPE32_)(_ATYPE_) /* widen if necessary */ +#define _ACAST64_ (_ATYPE64_) /* do _not_ narrow */ #endif
/*
could you fix this in the pcnet driver? If possible I want to avoid U-Boot specific changes on files imported from Linux.
--
- Daniel
Hi Daniel,
I've submitted an alternate fix in v2 of the series just now. It still requires a change to MIPS code but it mirrors one I have submitted to Linux, and hopefully one you'll find acceptable. It makes MIPS32 & MIPS64 more consistent & prevents drivers needing to care about which segment of the virtual address space an address is in when calling virt_to_phys.
Thanks, Paul