Re: (regression) kernel/timeconst.h bugs with HZ=128

From: H. Peter Anvin
Date: Tue Feb 26 2008 - 19:39:06 EST


Andrew Morton wrote:
On Tue, 26 Feb 2008 10:32:24 -0800 David Brownell <david-b@xxxxxxxxxxx> wrote:

I see these warnings on 32 bit ARM systems:

CC kernel/time.o
kernel/time.c: In function 'msecs_to_jiffies':
kernel/time.c:472: warning: integer constant is too large for 'long' type
kernel/time.c: In function 'usecs_to_jiffies':
kernel/time.c:487: warning: integer constant is too large for 'long' type

Line 472: return ((u64)MSEC_TO_HZ_MUL32 * m + MSEC_TO_HZ_ADJ32)
line 487:
return ((u64)USEC_TO_HZ_MUL32 * u + USEC_TO_HZ_ADJ32)

The problem seems to be that these constants from kernel/timeconst.h
have too many digits:

#define ONLY_THIRTYTWO_BITS 0x01234567

#define MSEC_TO_HZ_ADJ32 0x3f7ced916
#define USEC_TO_HZ_ADJ32 0xfffbce4217d

Those *_ADJ32 constants should have "ULL" suffixes, yes?
Adding that by hand resolves the problem, but only until
the next time that header file gets regenerated.

Someone with observable Perl-fu should fix this ...



<barf>

The real issue is that we don't have any equivalent to the UINTxx_C() creation macros in Linux, and even though we're using a cast -- which is totally correct -- makes gcc issue a (spurious) warning.

The right thing would be to create _C() macros and use those in the C code, i.e. UINT64_C(MSEC_TO_HZ_MUL32).

I'll write up a patch, but probably not today.

-hpa
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/