Article 4625 of alt.sys.pdp10: Path: news3.best.com!news1.best.com!144.212.95.13.MISMATCH!newsfeed.mathworks.com!nntprelay.mathworks.com!cam-news-hub1.bbnplanet.com!news.gtei.net!newsfeed.xcom.net!news.ultranet.com!not-for-mail From: "Alan H. Martin" Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Wed, 23 Dec 1998 15:38:39 -0500 Organization: UltraNet Communications , an RCN Company http://www.ultranet.com/ Lines: 23 Message-ID: <368154CF.3570F64F@MA.UltraNet.Com> References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> NNTP-Posting-Host: d20.dial-7.cmb.ma.ultra.net Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Complaints-To: abuse@ultra.net X-Ultra-Time: 23 Dec 1998 20:38:56 GMT X-Mailer: Mozilla 4.5 [en] (Win95; I) X-Accept-Language: en,en-US,en-GB,es Xref: news3.best.com alt.sys.pdp10:4625 Ivy wrote: > I just read my post , It sounds like I'm implying the XKL and Jupiter > had 22bit phys and 23bit virtual this is NOT TRUE > The XKL does and the Jupiter should have had something like 30Bit > phys/virtual (If I read it correctly XKL-1 is 33bit physical 30 bit > virtual using 4096 sections ) The KL did (and Jupiter would have) all EA calcs with 30 bits of precision while running with extended addressing support. Tops-20 on the KL supported 23 bits of virtual address space (32 sections of 256K words). I guess some JSY couldn't reach across sections. I don't recall the virtual address limits designed into the Jupiter hardware, although they were surely larger than 32 sections. At some point, Tops-20 on Jupiter might have paused at supporting 27 bits of virtual address space, because that required no interface changes in JSY that held page numbers in halfwords. I'm not aware of designs for Tops-20 that would have added 3 more bits, but if the product line had continued, it surely would have expanded to be as wide as the hardware allowed by now. /AHM -- Alan Howard Martin AMartin@MA.UltraNet.Com Article 4626 of alt.sys.pdp10: Path: news3.best.com!news1.best.com!144.212.95.13.MISMATCH!newsfeed.mathworks.com!nntprelay.mathworks.com!logbridge.uoregon.edu!news.u.washington.edu!Tomobiki-Cho.CAC.Washington.EDU!mrc From: Mark Crispin Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Wed, 23 Dec 1998 16:56:35 -0800 Organization: Networks & Distributed Computing Lines: 18 Message-ID: References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> <368154CF.3570F64F@MA.UltraNet.Com> NNTP-Posting-Host: tomobiki-cho.cac.washington.edu Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Trace: nntp1.u.washington.edu 914460998 23668 (None) 140.142.17.37 X-Complaints-To: help@cac.washington.edu NNTP-Posting-User: rded To: "Alan H. Martin" In-Reply-To: <368154CF.3570F64F@MA.UltraNet.Com> Xref: news3.best.com alt.sys.pdp10:4626 On Wed, 23 Dec 1998, Alan H. Martin wrote: > The KL did (and Jupiter would have) all EA calcs with 30 bits of precision > while running with extended addressing support. Tops-20 on the KL supported > 23 bits of virtual address space (32 sections of 256K words). I guess some > JSY couldn't reach across sections. No, it's because there were only 32 words in the EPT for the section table on the KL. I think that the way that Jupiter and XKL did it was by having an additional level of tables. Assuming that a section table would be in a page by itself, you'd need an 8 word table of section tables to cover all 30 bits. -- Mark -- * RCW 19.149 notice: This email address is located in Washington State. * * Unsolicited commercial email may be billed $500 per message. * Science does not emerge from voting, party politics, or public debate. Article 4627 of alt.sys.pdp10: Path: news3.best.com!news2.best.com!newshub.northeast.verio.net!newsfeed.cwix.com!18.24.4.11!newsswitch.lcs.mit.edu!news.ultranet.com!not-for-mail From: "Alan H. Martin" Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Wed, 23 Dec 1998 21:15:52 -0500 Organization: UltraNet Communications , an RCN Company http://www.ultranet.com/ Lines: 24 Message-ID: <3681A3D8.A6192E70@MA.UltraNet.Com> References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> <368154CF.3570F64F@MA.UltraNet.Com> NNTP-Posting-Host: d123.dial-5.cmb.ma.ultra.net Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Complaints-To: abuse@ultra.net X-Ultra-Time: 24 Dec 1998 02:16:09 GMT X-Mailer: Mozilla 4.5 [en] (Win95; I) X-Accept-Language: en,en-US,en-GB,es Xref: news3.best.com alt.sys.pdp10:4627 Mark Crispin wrote: > > On Wed, 23 Dec 1998, Alan H. Martin wrote: > > The KL did (and Jupiter would have) all EA calcs with 30 bits of precision > > while running with extended addressing support. Tops-20 on the KL supported > > 23 bits of virtual address space (32 sections of 256K words). I guess some > > JSY couldn't reach across sections. > > No, it's because there were only 32 words in the EPT for the section table > on the KL. I think that the way that Jupiter and XKL did it was by having > an additional level of tables. Assuming that a section table would be in > a page by itself, you'd need an 8 word table of section tables to cover > all 30 bits. The VM implementation data structures are not sufficient to address the issue. To rephrase: there existed FOO% JSY which were *not* reimplemented as XFOO% with fullword address field arguments and results. Hence, not every JSYS was capable of pointing to out-of-section arguments. Searching MONSYM for ``777777'' uncovers examples like CM%GJB in .CMGJB for COMND%. Those JSY that took an address in an 18-bit field couldn't reach across sections. /AHM -- Alan Howard Martin AMartin@MA.UltraNet.Com Article 4630 of alt.sys.pdp10: Path: news3.best.com!news1.best.com!su-news-hub1.bbnplanet.com!news.gtei.net!newsfeed.berkeley.edu!news.u.washington.edu!Tomobiki-Cho.CAC.Washington.EDU!mrc From: Mark Crispin Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Thu, 24 Dec 1998 00:02:18 -0800 Organization: Networks & Distributed Computing Lines: 42 Message-ID: References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> <368154CF.3570F64F@MA.UltraNet.Com> <3681A3D8.A6192E70@MA.UltraNet.Com> NNTP-Posting-Host: tomobiki-cho.cac.washington.edu Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Trace: nntp1.u.washington.edu 914486540 36022 (None) 140.142.17.39 X-Complaints-To: help@cac.washington.edu NNTP-Posting-User: ajohnson To: "Alan H. Martin" In-Reply-To: <3681A3D8.A6192E70@MA.UltraNet.Com> Xref: news3.best.com alt.sys.pdp10:4630 On Wed, 23 Dec 1998, Alan H. Martin wrote: > The VM implementation data structures are not sufficient to address the issue. Yes, but you were saying that the KL supported 30 bit addressing which it clearly did not. The KL was limited to 22 bits of hardware addressed memory and 23 bits of virtual addressed memory, and these were hardware and microcode restrictions, not operating system restrictions (although you would need to make OS changes to support a different organization of section tables). > To rephrase: there existed FOO% JSY which were *not* reimplemented as XFOO% > with fullword address field arguments and results. Hence, not every JSYS was > capable of pointing to out-of-section arguments. Searching MONSYM for > ``777777'' uncovers examples like CM%GJB in .CMGJB for COMND%. Those JSY that > took an address in an 18-bit field couldn't reach across sections. This actually wasn't particularly important. I am one of the few people who has actually written a multi-section program from scratch, as opposed to converting an existing single-section program. [Quick, anyone know what it was?] Most memory references were in-section, and this included JSYS arguments. Most of what you needed to reference in other sections were things that you'd access with byte pointers, and the introduction of OWGBPs greatly reduced the need to add a bunch of new jsi. At least on the KL, you wanted to be careful about the sorts of memory referencess you did inter-section, since you could really screw up performance big time (Chuck Hedrick found that out with CLISP!) if you are now aware of how the hardware page table (which caches virtual to physical references in the microcode so it doesn't have to go through the section and page tables each time). The problem was that if you run multi-section, you can have multiple pages which use the same HPT entry, so if you reference a page whose HPT entry is in use by another page, the HPT gets invalidated. I forget if it just does a CLRPT on that one HPT entry or if it effectively does a DATAO PAG, (invalidating the entire HPT); I seem to recall that it was the latter. -- Mark -- * RCW 19.149 notice: This email address is located in Washington State. * * Unsolicited commercial email may be billed $500 per message. * Science does not emerge from voting, party politics, or public debate. Article 4631 of alt.sys.pdp10: Path: news3.best.com!nntp1.ba.best.com!not-for-mail From: inwap@best.com (Smith and O'Halloran) Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: 24 Dec 1998 01:52:20 -0800 Organization: a user of Best Internet Communications, Inc. www.best.com Message-ID: <75t2sk$aul$1@shell3.ba.best.com> References: <36794982.265795064@netnews.worldnet.att.net> <3681A3D8.A6192E70@MA.UltraNet.Com> Lines: 16 NNTP-Posting-Host: shell3.ba.best.com X-Trace: nntp1.ba.best.com 914493144 9744 inwap@206.184.139.134 Xref: news3.best.com alt.sys.pdp10:4631 In article , Mark Crispin wrote: >Yes, but you were saying that the KL supported 30 bit addressing which it >clearly did not. The KL was limited to 23 bits of virtual address ... Minor quibble: The KL used 30 bit addressing, but the high-order 7 bits were a MBZ field. (Must Be Zero.) That is, if the virtual address was too big to fit into 23 bits, you would get a page fault. The high-order 7 bits were not ignored, which would have been the case for modulo 23 bit addressing. -Joe -- INWAP.COM is Joe Smith, Sally Smith and our cat Murdock. (The O'Hallorans and their cats moved to http://www.tyedye.org/ Nov-98.) See http://www.inwap.com/ for PDP-10, "ReBoot", "Shadow Raiders"/"War Planets" Article 4637 of alt.sys.pdp10: Path: news3.best.com!news2.best.com!news.maxwell.syr.edu!netnews.com!chnws02.mediaone.net!24.131.1.12!denws01.mw.mediaone.net!news.altair.com!NewsNG.Chicago.Qual.Net!news.indiana.edu!news.ind.net!news.iupui.edu!haystack!mhwood From: mhwood@Ameritech.net (Mark H. Wood) Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Fri, 25 Dec 1998 15:43:30 GMT Organization: La Petite Hackerie Lines: 26 Message-ID: References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> <368154CF.3570F64F@MA.UltraNet.Com> <3681A3D8.A6192E70@MA.UltraNet.Com> NNTP-Posting-Host: mhw.ulib.iupui.edu X-Newsreader: TIN [UNIX 1.3 950515BETA PL0] Xref: news3.best.com alt.sys.pdp10:4637 Mark Crispin (mrc@CAC.Washington.EDU) wrote: [snip] : This actually wasn't particularly important. I am one of the few people : who has actually written a multi-section program from scratch, as opposed : to converting an existing single-section program. [Quick, anyone know : what it was?] Most memory references were in-section, and this included : JSYS arguments. Most of what you needed to reference in other sections : were things that you'd access with byte pointers, and the introduction of : OWGBPs greatly reduced the need to add a bunch of new jsi. It's been a long time, but I can second this, as I also wrote *one* multi-section program from scratch. (It was a sickly imitation of the DECmail-MS UA, without the network stuff, done before I heard about MM. I PMAP%ed the mailbox into an otherwise empty section for speed and ease of manipulation at the expense of putting hard limits on mailbox size. It was really fast, and lots of fun to build, but that's about the most I can say for it.) The quote above pretty accurately reflects my experience, what I can remember of it after a decade or so. (Or did you mean multiple *code* sections? I've not done that.) -- -- Mark H. Wood, radical centrist mhwood@ameritech.net Charlie, put down that Glitter Glue -- it's time to show the audience some content! Article 4638 of alt.sys.pdp10: Path: news3.best.com!news2.best.com!news.maxwell.syr.edu!news.idt.net!nntp.farm.idt.net!news From: "Chris Ward" Newsgroups: alt.sys.pdp10 Subject: Re: The True Faith Date: Sun, 27 Dec 1998 11:25:51 -0500 Organization: IDT (Best News In The World) Lines: 28 Message-ID: <765n5h$o1d@nnrp3.farm.idt.net> References: <36794982.265795064@netnews.worldnet.att.net> <7591ch$kks@bonkers.taronga.com> <75amfq$cqt$1@ligarius.ultra.net> <36796950.0@news.wizvax.net> <75dkbn$2i2@nnrp4.farm.idt.net> <75dqop$pqe$1@ligarius.ultra.net> <75eplu$2ne@nnrp2.farm.idt.net> <367B0ED6.238FD95D@s054.aone.net.au> <367B11FF.73D40384@s054.aone.net.au> <368154CF.3570F64F@MA.UltraNet.Com> <3681A3D8.A6192E70@MA.UltraNet.Com> NNTP-Posting-Host: ppp-28.ts-1-bay.hob.idt.net X-Newsreader: Microsoft Outlook Express 4.72.3115.0 X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3 Xref: news3.best.com alt.sys.pdp10:4638 Mark H. Wood wrote in message ... >It's been a long time, but I can second this, as I also wrote *one* >multi-section program from scratch Of course, once you start doing multi-segment programs, you start kicking at one of the the things that made the PDP-10 instruction set great - It was relatively easy to understand, and mere mortals (like myself) could get work done in it. Fortunately, 256k words (~1mb) was enough to write a lot of code in, even if most of it was Fortran. And I say most realistically. There were many features of the 10, or of any machine that you really wanted to "get at", but you could not unless you used some assembly. The learning curve, I think for competing machines like the 370 was a lot longer for assembly, and full utilisation of the machine was a lot more difficult. In other words, on the Dec-10 you could get undergraduate students to work "wonders" because the engineers at DEC had built a machine that had the needed functions, with good performance (better on a dollar basis!), are presented them in an organized manner. What I am saying is that the 18 bit length of the E calculation was in many ways the crux of the machine, and had the PDP-10 been expanded, my inclination would have to have build a 72-bit machine (or some such) with an E calculation of 36 bits. Try to build a culturally compatible machine, and try not to add too many instructions like the Vax did. Article 8055 of alt.sys.pdp10: Sender: eric@ruckus.brouhaha.com From: Eric Smith Newsgroups: alt.sys.pdp10 Subject: Re: ITS CD (was Re: Wanna see how to annoy MRC? ^_^) References: <8bm5iu$prp$1@saltmine.radix.net> <8bmjph$mps$1@199.201.191.2> <38E6B5EA.5B112B79@bellatlantic.net> <8c7smv$jra$1@saltmine.radix.net> X-Disclaimer: Everything I write is false. Organization: Eric Conspiracy Secret Labs X-Eric-Conspiracy: There is no conspiracy. Date: 11 Apr 2000 15:00:29 -0700 Message-ID: Lines: 16 X-Newsreader: Gnus v5.5/Emacs 20.3 NNTP-Posting-Host: ruckus.brouhaha.com X-Trace: 11 Apr 2000 15:57:32 -0800, ruckus.brouhaha.com Path: news3.best.com!news2.best.com!newsfeed.berkeley.edu!enews.sgi.com!news.sgi.com!news.spies.com!ruckus.brouhaha.com Xref: news3.best.com alt.sys.pdp10:8055 Mark Crispin writes: > Needless to say, TOPS-20 style virtual memory is probably unimplementable > on modern CPUs. I doubt that very much, for two reasons: 1) Much of the TOPS-20 development was done on KI10s, since KL10s weren't yet available. As you're well aware, the KI10 does not support any of the fancy paging features used by "TOPS-20 Paging" on the KL10. 2) Many modern CPUs (e.g., MIPS) provide a TLB but leave management of the TLB entirely to software. Such CPUs do not dictate the format (or even the existence) of page tables. If there's some specific aspect of "TOPS-20 style virtual memory" which you believe is unimplementable on modern CPUs, please explain. Article 8060 of alt.sys.pdp10: Path: news3.best.com!news2.best.com!feed1.news.rcn.net!rcn!chnws02.mediaone.net!chnws05.ne.mediaone.net!24.128.8.70!typhoon.ne.mediaone.net.POSTED!not-for-mail Newsgroups: alt.sys.pdp10 Subject: Re: ITS CD (was Re: Wanna see how to annoy MRC? ^_^) References: <8bm5iu$prp$1@saltmine.radix.net> <8bmjph$mps$1@199.201.191.2> <38E6B5EA.5B112B79@bellatlantic.net> <8c7smv$jra$1@saltmine.radix.net> From: Ric Werme X-Newsreader: NN version 6.5.0 CURRENT #119 Lines: 28 Message-ID: Date: Wed, 12 Apr 2000 01:37:20 GMT NNTP-Posting-Host: 24.218.4.112 X-Complaints-To: abuse@mediaone.net X-Trace: typhoon.ne.mediaone.net 955503440 24.218.4.112 (Tue, 11 Apr 2000 21:37:20 EDT) NNTP-Posting-Date: Tue, 11 Apr 2000 21:37:20 EDT Organization: Road Runner Xref: news3.best.com alt.sys.pdp10:8060 Eric Smith writes: >Mark Crispin writes: >> Needless to say, TOPS-20 style virtual memory is probably unimplementable >> on modern CPUs. >I doubt that very much, for two reasons: >1) Much of the TOPS-20 development was done on KI10s, since KL10s weren't > yet available. As you're well aware, the KI10 does not support any > of the fancy paging features used by "TOPS-20 Paging" on the KL10. >2) Many modern CPUs (e.g., MIPS) provide a TLB but leave management of > the TLB entirely to software. Such CPUs do not dictate the format > (or even the existence) of page tables. Page faults on Alpha trap to "PAL code" which is just Alpha code that has a lot of restrictions and lot of extra abilities. I've long since forgotten most of TENEX/TOPS-20 paging did, but I think that some of the LRU/aging stuff wouldn't put much load on Alpha PAL code. A lot of fundamental stuff is done in PAL code, HALT, interrupts, TLB maintenance, system calls entry, etc. -Ric -- Ric Werme | werme@nospam.mediaone.net http://people.ne.mediaone.net/werme | ^^^^^^^ delete