*BSD News Article 6882


Return to BSD News archive

Newsgroups: comp.unix.bsd
Path: sserve!manuel.anu.edu.au!munnari.oz.au!sgiblab!zaphod.mps.ohio-state.edu!cs.utexas.edu!uunet!super!rminnich
From: rminnich@metropolis.super.org (Ronald G Minnich)
Subject: Re: Adding Swapspace ??
Message-ID: <1992Oct21.124610.14306@super.org>
Sender: rminnich@metropolis (Ronald G Minnich)
Nntp-Posting-Host: metropolis
Organization: Supercomputing Research Center
References: <Bw7H4L.LLB@cosy.sbg.ac.at> <1992Oct16.162729.3701@ninja.zso.dec.com> <1992Oct16.201806.21519@fcom.cc.utah.edu> <17072@ksr.com>
Date: Wed, 21 Oct 1992 12:46:10 GMT
Lines: 36

In article <17072@ksr.com>, dean@ksr.com (Dean Anderson) writes:
|> Do mach external pagers work in 386BSD?  Will they work in 4.4BSD?

well, don't know about 386bsd. I do know that an external pager i have had running
on SunOS for the past few years (now implemented using NFS as a base) can't be 
made to work on 386BSD without some simple changes in the VM code. I am not 
sure how you would get one working given the way the code is now, unless you
really want to have a custom version of vm_fault for your pager. 

In the VM code on 386BSD, the fault code figures out what to do, and then 
asks the lower levels to get a page, and gives the lower levels an object pager
and a page, viz:
      rv = vm_pager_get(object->pager, m, TRUE)
The 
problem is that the information delivered by vm_fault to lower levels is 
insufficient in at least the following ways:
1) no info on faulting virtual address
2) no info on faulting protection type 

note that this is trivial to add, so i guess it is not done because information 
hiding is good, or something. I only know that my inquiries to a 4.4bsd person
about adding this went nowhere. 

So i will at some point be making this available, but it will require some 
non-blessed hacks to the vm code (albeit trivial) starting from vm_fault 
on down. Not a high priority here, as we have at least 200 SunOS machines and 
maybe 3 386BSD machines. 

For what it is worth, i much prefer the SunOS vm scheme to what i have seen
of the Mach scheme. I have been able to do a lot of interesting work over the 
past few years by exploiting the SunOS VM architecture. If you have not read the
papers on the SunOS VM, you should. 
-- 
Never assume a conspiracy when you can assume     | rminnich@super.org
incompetence.                                     | (301)-805-7451 or 7312
Most abused words: open (can you believe IBM calls AIX open?), scaleable