Re: linux-next: build warning after merge of the nvmem tree

["H. Nikolaus Schaller" <hns@xxxxxxxxxxxxx>]

Hi,

> Am 06.03.2020 um 05:28 schrieb Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>:
> 
> Hi all,
> 
> After merging the nvmem tree, today's linux-next build (x86_64
> allmodconfig) produced this warning:
> 
> In file included from include/linux/clk.h:13,
>                 from drivers/nvmem/jz4780-efuse.c:25:
> drivers/nvmem/jz4780-efuse.c: In function 'jz4780_efuse_read':
> include/linux/kernel.h:842:29: warning: comparison of distinct pointer types lacks a cast
>  842 |   (!!(sizeof((typeof(x) *)1 == (typeof(y) *)1)))
>      |                             ^~
> include/linux/kernel.h:856:4: note: in expansion of macro '__typecheck'
>  856 |   (__typecheck(x, y) && __no_side_effects(x, y))
>      |    ^~~~~~~~~~~
> include/linux/kernel.h:866:24: note: in expansion of macro '__safe_cmp'
>  866 |  __builtin_choose_expr(__safe_cmp(x, y), \
>      |                        ^~~~~~~~~~
> include/linux/kernel.h:875:19: note: in expansion of macro '__careful_cmp'
>  875 | #define min(x, y) __careful_cmp(x, y, <)
>      |                   ^~~~~~~~~~~~~
> drivers/nvmem/jz4780-efuse.c:76:24: note: in expansion of macro 'min'
>   76 |   unsigned int chunk = min(bytes, (start + JZ_EFU_READ_SIZE)
>      |                        ^~~
> 
> Introduced by commit
> 
>  50a09dfd394a ("nvmem: add driver for JZ4780 efuse")

The new kbuild robot message is more precise.

drivers/nvmem/jz4780-efuse.c:76:38: sparse: sparse: incompatible types in comparison expression (different type sizes):
drivers/nvmem/jz4780-efuse.c:76:38: sparse:    unsigned long *
drivers/nvmem/jz4780-efuse.c:76:38: sparse:    unsigned int *

A look into the code:

/* main entry point */
static int jz4780_efuse_read(void *context, unsigned int offset,
			     void *val, size_t bytes)
{
	struct jz4780_efuse *efuse = context;

	while (bytes > 0) {
		unsigned int start = offset & ~(JZ_EFU_READ_SIZE - 1);
		unsigned int chunk = min(bytes, (start + JZ_EFU_READ_SIZE)
					 - offset);

The problem seems to be that size_t is not always compatible to
unsigned int on all machines and compilers. My cross-gcc for MIPS
did not complain. kbuild robot tried for ARCH=x86_64 and in the
first warning for GCC_VERSION=7.5.0 make.cross ARCH=riscv 

So I think we have to use

		size_t start = offset & ~(JZ_EFU_READ_SIZE - 1);
		size_t chunk = min(bytes, (start + JZ_EFU_READ_SIZE)
					 - offset);

I'l send a patch (after trying on MIPS).

BR and thanks,
Nikolaus