Re: [PATCH] nodemask_t x86_64 changes [5/7]

[Keith Owens <kaos@sgi.com>]

On Thu, 25 Mar 2004 23:14:12 -0800, 
Paul Jackson <pj@sgi.com> wrote:
>Does your way work if NR_CPUS is less than BITS_PER_LONG?
>Won't gcc complain upon seeing something like, for say
>NR_CPUS = 4 on a 32 bit system:
>
>   { [ 0 ... -1 ] = ~0UL, ~0UL << 28 }
>
>with the errors and warnings:
>
>  error: empty index range in initializer
>  warning: excess elements in struct initializer

I only did one case, to concentrate on the value for the last word.  A
full implementation has to cater for NR_CPUS < BITS_PER_LONG as well.

>and shouldn't the last word be inverted: ~(~0UL << NR_CPUS_UNDEF) ?

For big endian, ~0UL << NR_CPUS_UNDEF is right.  For little endian, it
depends on how you represent an incomplete bit map.  Is it represented
as a pure bit string, i.e. as if the arch were big endian?  Or is it
represented as a mapping onto the bytes of the underlying long?