On Wed, 11 Jul 2007 15:28:19 +0100 (BST), "Maciej W. Rozycki" <macro@xxxxxxxxxxxxxx> wrote: > It looks like a bug in sparse. The result of CKSEG1ADDR() has the same > size as the pointer. Perhaps we could append 'L' to the expansion of > KSEG1 et al, but that should not really matter. Yes, adding 'L' to KSEG1 is another way to silence the warnings. But I just thought it was a bit intrusive. And I'm not sure all code are OK if KSEG1 is 'signed' ... > But -- I have just checked two example calls to this function, one with a > 32-bit configuration and another one with a 64-bit one and sparse did not > complain. The cpp expansions of the expression in question are: > > return (void *)((((int)(int)(phys_addr)) & 0x1fffffff) | 0xa0000000); > > and: > > return (void *)((((long int)(int)(phys_addr)) & 0x1fffffff) | 0xffffffffa0000000L); > > respectively, so your cast is definitely redundant in these cases. What > sort of configuration are you using? What's the preprocessor output for > the problematic case? I see the warnings on 32-bit qemu kernel. drivers/serial/8250.c, lib/devres.c, etc. I think sparse complains on casting to "void __iomem *" from "int". It looks sparse accepts casting from "long". --- Atsushi Nemoto
