lynn@booboo.SanDiego.NCR.COM (Lynn Hedegard) (06/29/89)
I am working in a kernel group which is porting AT&T's V.4 to our new hardware. I am trying to understand the new Virt Mem architecture in V.4 and would appretiate any recomendation regarding articles which describe this topic. I am looking for papers which describe the data structures and algorithms used.
guy@auspex.auspex.com (Guy Harris) (07/01/89)
>I am working in a kernel group which is porting AT&T's V.4 to our >new hardware. I am trying to understand the new Virt Mem >architecture in V.4 and would appretiate any recomendation >regarding articles which describe this topic. I am looking for >papers which describe the data structures and algorithms used. Try: *Virtual Memory Architecture in SunOS*, from the Summer 1987 USENIX proceedings *SunOS Virtual Memory Implementation*, from either the most recent EUUG proceedings or the one before that, I think (V.4's VM implementation is derived from the SunOS 4.0 one, although there are differences in a few details). They aren't extremely detailed, but they do cover a fair bit at a relatively high level. With any luck, all you'll have to do is write a "hat (Hardware Address Translation) layer" for your MMU, which'll just plug into the rest of the VM subsystem. (If the new hardware uses a 68020+PMMU, or a 68030 or so - i.e., if it's using a Motorola-style MMU - you might just see if Sun'll sell you their 4.0 implementation; I have no idea if they will, but it's worth a try....) If you *do* have to do more, let AT&T (and possibly Sun) know, so that with any luck the implementation can be modified to let the new "hat layer" plug into it in the future. While you're at it, point the languages people at *Shared Libraries in SunOS* in the Summer 1987 USENIX proceedings for some information on how the shared library mechanism works in V.4....
greg@cheers.UUCP (Greg Onufer) (07/04/89)
lynn@booboo.SanDiego.NCR.COM (Lynn Hedegard) writes: >I am working in a kernel group which is porting AT&T's V.4 to our >new hardware. I am trying to understand the new Virt Mem >architecture in V.4 and would appretiate any recomendation "Virtual Memory Architecture in SunOS" Robert A. Gingell Joseph P. Moran William A. Shannon Sun Microsystems, Inc. 2550 Garcia Ave Mountain View, CA 94043 This describes the new VM Architecture developed by Sun for SunOS and adopted by AT&T for ATT System V.4 (As far as I know). Cheers!greg
hiho@cbnewsl.ATT.COM (paul.ho) (07/04/89)
> With any luck, all you'll have to do is write a "hat (Hardware Address > Translation) layer" for your MMU, which'll just plug into the rest of > the VM subsystem. (If the new hardware uses a 68020+PMMU, or a 68030 or > so - i.e., if it's using a Motorola-style MMU - you might just see if > Sun'll sell you their 4.0 implementation; I have no idea if they will, > but it's worth a try....) If you *do* have to do more, let AT&T (and > possibly Sun) know, so that with any luck the implementation can be > modified to let the new "hat layer" plug into it in the future. Is region (introduced in SVR3.0) dead? Or does it co-exist with the VM stuff in SVR4? Thanks.
gwyn@smoke.BRL.MIL (Doug Gwyn) (07/07/89)
In article <987@cbnewsl.ATT.COM> hiho@cbnewsl.ATT.COM (paul.ho) writes: >Is region (introduced in SVR3.0) dead? Or does it co-exist with >the VM stuff in SVR4? Thanks. From the AT&T BOF at USENIX recently, I'd say that the region approach was totally replaced by VM. The implication is that SVR4 will be a real bitch to port to a system that doesn't have an MMU that conforms to the VM model. E.g. some supercomputers..
prc@erbe.se (Robert Claeson) (07/10/89)
In article <10499@smoke.BRL.MIL> gwyn@brl.arpa (Doug Gwyn) writes: >From the AT&T BOF at USENIX recently, I'd say that the region approach >was totally replaced by VM. The implication is that SVR4 will be a >real bitch to port to a system that doesn't have an MMU that conforms >to the VM model. E.g. some supercomputers.. And I guess that we will never see SVR4 (or even SVR3) ported to the intel 80286 chip... -- Robert Claeson E-mail: rclaeson@erbe.se ERBE DATA AB
dwc@cbnewsh.ATT.COM (Malaclypse the Elder) (07/11/89)
In article <10499@smoke.BRL.MIL>, gwyn@smoke.BRL.MIL (Doug Gwyn) writes: > From the AT&T BOF at USENIX recently, I'd say that the region approach > was totally replaced by VM. The implication is that SVR4 will be a > real bitch to port to a system that doesn't have an MMU that conforms > to the VM model. E.g. some supercomputers.. from what i have seen of the code, at&t's VM code should not be too hard to port to non-VM machines. you just have to make sure that you will never take a page fault and then the hat routines will mostly be no-ops. . then you just make all the segment drivers no-ops (or define new segment types). of course, you'll lose all the nice memory mapping (e.g. mmap()) system calls. danny chen att!hocus!dwc