On Thu, Sep 26, 2002 at 05:46:36PM -0400,  wrote:
> Hmm I wanted to send that one to codalist, but forgot to change the
> address, you'll see the same message show up soon ;)
 
Today is one of those days that nothing seems to go right, ever have that?

On Thu, Sep 26, 2002 at 10:49:22PM +0200, Ivan Popov wrote:
> > release_segment unmap failed RVM_ENOT_MAPPED  [this is what warries
> > me]
> 
> It looks like I have to play with rvm parameters until I fit that 1G
> into
> the virtual memory without collisions... Is somebody using 1G rvm on
> Linux?
> 
> My numbers are
> dsparm=1073741824
> heapsize=0x3FF00000
> rvmstart=0x50000000
> staticsize=0x100000
> nlists=80
> chunk=32
>         
> are there other ones, in use and not colliding?

You can look at the memory layout of a smaller server in /proc

 cat /proc/`pidof codasrv`/maps

There should be about 1.7GB of addressable space between 0x50000000 and
0xbfffc000.

But looking at your other numbers, heapsize = 0x3ff0 and static size is
0x10, which adds up to exactly dsparm (0x4000). But RVM needs at least
one page for the header, and RDS needs one page for the list heads of
the allocation chains.

So you could reduce heapsize (or static size) a bit and things should
fit. 0x3fefd000 should already leave 2 pages free, but maybe 0x3fef0000
is safer, and only wastes about 50KB.

Jan