Jason Thorpe <thorpej@nas.nasa.gov> writes:

> ...
> This is all handled in the kernel... there is no hoakey kernel-userland
> protocol to deal with... nice, clean, simple.
> 
> Am I making any sense? :-)
Cool ...
Note. In Coda, we need to start a server to get mount to work so I'd
prefer to let it to the server load the Coda module vs a mount of
/coda.

> 
>  > - /dev/ksyms, to be able to access an up-to-date symbol table. (as a side
>  >   effect ps stopped to care about from which kernel the system was loaded :-)
> 
Ignoring the issue of in what directory this service should reside,
I'd also like to see a "/dev/nl".  If I need namelist data, I'd like
to query the kernel exactly what I need and not the entire symbol table.

Finally, I'd also like to suggest that not all the symbols in the
kernel should be accessible.  Imagine, for example, that some company
writes a lkm that knows about the internal details of vm and then we
change to uvm.  Perhaps, we could take the union of all the external
references in the existing lkm's and use this as an initial set that
defines a kernel service API.  Assuming this set is small, we can 
export the symbols directly from the kernel w/o significant bloat.