msync on Hurd, dev_t on MIPS

[Russ Allbery]

Julien ÉLIE <julien at trigofacile.com> writes:

>> I wonder if MMAP_NEEDS_MSYNC or MMAP_MISSES_WRITES are defined on Hurd.
>> If so, this error message means that there will probably be some data
>> corruption if someone runs innd on the Hurd.

> I believe both of these variables are undefined on the Hurd:

> https://buildd.debian.org/fetch.cgi?pkg=inn2;ver=2.5.1-1;arch=hurd-i386;stamp=1256745009

> Automatic build of inn2_2.5.1-1 on rossini by sbuild/hurd-i386 98-farm

> checking for working mmap... yes
> checking for madvise... no
> checking whether mmap sees writes... yes
> checking whether msync is needed... no
> checking how many arguments msync takes... 3

Interesting.  In that case, we should probably test for a working msync in
configure and, if msync returns a non-zero exit status in a simple test,
just stub out all of the calls with a #define to /* empty */.  What we're
doing now is functionally equivalent, so not a rush.

> So finally, SysV shared memory on the Hurd and dev_t on MIPS do not seem
> to be properly implemented and we cannot do anything for that.

The dev_t thing on MIPS is very odd.  I wouldn't expect that sort of
difference between different Linux architectures.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.