Re: do_div vs sector_t

From: Neil Brown <neilb@cse.unsw.edu.au>
To: Matthew Wilcox <willy@debian.org>
Cc: Bernardo Innocenti <bernie@develer.com>,
	Andrew Morton <akpm@zip.com.au>,
	linux-kernel@vger.kernel.org
Subject: Re: do_div vs sector_t
Date: Sat, 12 Jul 2003 08:42:18 +1000	[thread overview]
Message-ID: <16143.15690.106092.770785@gargle.gargle.HOWL> (raw)
In-Reply-To: message from Matthew Wilcox on Friday July 11

On Friday July 11, willy@debian.org wrote:
> 
> # define do_div(n,base) ({                              \
>         uint32_t __base = (base);                       \
>         uint32_t __rem;                                 \
>         if (likely(((n) >> 32) == 0)) {                 \
> 
> so if we call do_div() on a u32, the compiler emits nasal daemons.
> and we do this -- in the antcipatory scheduler:
> 
>                 if (aic->seek_samples) {
>                         aic->seek_mean = aic->seek_total + 128;
>                         do_div(aic->seek_mean, aic->seek_samples);
>                 }
> 
> seek_mean is a sector_t so sometimes it's 64-bit on a 32-bit platform.
> so we can't avoid calling do_div().
> 
> This almost works (the warning is harmless since gcc optimises away the call)
> 
> # define do_div(n,base) ({                                              \
>         uint32_t __base = (base);                                       \
>         uint32_t __rem;                                                 \
>         if ((sizeof(n) < 8) || (likely(((n) >> 32) == 0))) {            \
>                 __rem = (uint32_t)(n) % __base;                         \
>                 (n) = (uint32_t)(n) / __base;                           \
>         } else                                                          \
>                 __rem = __div64_32(&(n), __base);                       \
>         __rem;                                                          \
>  })
> 
> Better ideas?

sector_div, defined in blkdev.h, is probably what you want.

NeilBrown