On Tue, Jul 24, 2007 at 07:13:23PM +0200, Joerg Sonnenberger wrote:
> On Tue, Jul 24, 2007 at 07:06:04PM +0200, Matthias Drochner wrote:
> > NetBSD's new mremap(2) does not only require the start address
> > of the memory range to be page aligned, but additionally the
> > old and new sizes to be multiple of the page size.
> > This differs from the Linux behaviour, and it is also
> > inconsistent to mmap(2) which rounds up implicitely.
> 
> I always found the mmap behaviour a bit bogus, but I don't have a strong
> position on it. Feel free to commit.

hmmm... beyond a bit bogus, the behaviour of our mmap just seems wrong:
 if you map a 10 byte file (for example), the mmap man page says that
the mapped region may be extended up to the page size.  That seems
reasonable to me, but the man page also says any bytes beyond the end
of the mapped object will be zero filled.  This isn't always the case.
When I try it I get data that was in memory from a previous run of my
program.  (char *p=mmap(), assert(p[11] == 0), p[11] = 'X' fails on the
second run)
  I suppose there's a bit of a performance hit with zeroing out the page
every time, but it seems like a necessary thing to do.  A couple other
OSes I've tried it on do so.

eric