From: Digestifier To: Linux-Development@senator-bedfellow.mit.edu Reply-To: Linux-Development@senator-bedfellow.mit.edu Date: Tue, 13 Sep 94 07:13:09 EDT Subject: Linux-Development Digest #165 Linux-Development Digest #165, Volume #2 Tue, 13 Sep 94 07:13:09 EDT Contents: Games Make the OS (was: Re: Can't Run Doom!!) (Pete Deuel) LOOPBACK not handling packet load. (Chip Edwards) Re: Future of linux -- the sequel (Shannon Hendrix) Re: Not identifying ST-506 drives (was: Re: IDE Hard Drives w/ over 1024 cylinders) (Mark Lord) Re: Multiprocessing Pentium Systems (Hugh Emberson) Re: 3c509 Problems (Alan Modra) Re: 3c509 Problems (Brian Kramer) Interested in EIDE driver ? (Petri T J Mattila) Compiler benchmarking. was Survey: who wants f77,cc,c++,hpf for linux? (Robert Lipe) Re: Alpha Linux (Maarten Boekhold (Who'd you expect??)) Re: Alpha Linux (Maarten Boekhold (Who'd you expect??)) Re: Acid (Joachim Schrod) Re: Survey: who wants f77,cc,c++,hpf for linux? (Alan Cox) ---------------------------------------------------------------------------- From: deuelpm@craft.camp.clarkson.edu (Pete Deuel) Crossposted-To: comp.os.linux.misc Subject: Games Make the OS (was: Re: Can't Run Doom!!) Date: Tue, 13 Sep 1994 00:38:16 GMT In article <351ms1$ccu@bruce.uncg.edu> root@tao.binary9.com (Nicholas J. Leon) writes: >When I try to run Doom on my Linux box I get the following error >after I choose which scenerio I want to play: >U_GetNumForName: SSTMINUS not found >and then proceeds to exit. >Does anyone know what I can do about this? You see! If you really want a popular OS, you really have to have some good games. Never have I seen so many learn so much about linux all at once! How many times do you think I was asked "Whass, uhh, I-P-uh-X?" when Doom came out for DOS/net (was it IPX? I can't recall--I may be blocking this!) Now, just to make it worse: Doom needs sound support... Oh, I've heard the machine guns and music and explosions galore, but I mean VOICE! That's right, when I'm wandering around the Doom-scape, I wanna have a Mic plugged in to my PAS or SB and be able to beg for my life, or try to form alliances, or the like... Jus' somethin' to chew on... Pete =================================================== "Actually, I'm a lab mouse on stilts..." E-mail: deuelpm@craft.camp.clarkson.edu =================================================== ------------------------------ From: gt6444c@prism.gatech.edu (Chip Edwards) Subject: LOOPBACK not handling packet load. Date: 10 Sep 1994 01:15:10 -0400 I'm not sure who to send this to soo.. Some have noticed the ping -f on the loopback device fails miserably. I have noticed now that certain programs which use udp on the loopback port have MAJOR packet loss. Can someone PLZ PLZ PLEEAZZ fix the driver. Or send me a patch. I would most gratiously appreciate it. Chip gt6444c@prism.gatech.edu -- Chip Edwards (gt6444c@prism.gatech.edu) Georgia Institute of Technology, Atlanta Georgia, 30332 uucp: ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!gt6444c Internet: gt6444c@prism.gatech.edu ------------------------------ From: shendrix@escape.widomaker.com (Shannon Hendrix) Subject: Re: Future of linux -- the sequel Date: Sun, 11 Sep 1994 23:23:22 GMT djohnson@seuss.ucsd.edu (Darin Johnson) writes: >A Sun *2*? The one they stopped supported in the mid-80's? >It had a 68010 in it! Maybe you mean a Sparcstation 2? Oops... >So then, compare the DMA speed of your ISA system to a Sparcstation >(or even a lowly VME system like a Sun 3). There's more to a >computer's performance than cpu speed. No kidding. But I wasn't talking about overall performance either. No doubt, DMA speed is better on *SOME* Sun's. However, you'll note that I said *INTEGER* performance, not DMA speed. Also, I've benchmarked my simple IDE drive in common tasks and throughput against the Sun's at work and I usually do better on my 486. The IDE interface is not good for multiple drives but most of my accesses come from one drive. IDE is 16-bits, SCSI is 8 so unless you have multple drives and put them under a severe load, the Sun's I use are slightly slower (a lot slower in some cases). Also, with VLB or PCI you can get SCSI cards that will outrun most of the drives anyway. Anyway, the day I can afford SCSI fast-wide I'll leave IDE behind. But for now IDE is my only choice and it's giving a lot of 8-bit SCSI systems a run for their money in the configuration I have. When I start making heavy use of the second drive IDE will be hurting me but there are also IDE interfaces out there now that let both drives talk at once just like SCSI. I'm thinking of getting one because a friend of mine got one and it improved multiple drive accesses quite a bit. No idea how though unless it just lies to the drives so they don't know the other master/slave is doing I/O. I remember that each drive was placed on it's own cable. >-- >Darin Johnson >djohnson@ucsd.edu > - I'm not a well adjusted person, but I play one on the net. -- csh =========================================================================== shendrix@escape.widomaker.com | Linux and BSD ------------------------------ From: mlord@bnr.ca (Mark Lord) Subject: Re: Not identifying ST-506 drives (was: Re: IDE Hard Drives w/ over 1024 cylinders) Date: 12 Sep 1994 19:31:12 GMT In article pe1chl@rabo.nl writes: >In <9409091592@fangorn> Michael Haardt <(michael)u31b3hs@pool.informatik.rwth-aachen.de> writes: > >>davism@latcs2.lat.oz.au (Mitch Davis) writes: > >> if (stat & (BUSY_STAT|ERR_STAT)) >>! printk ("hd%c: ST506 interface, %dMB, CHS=%d/%d/%d\n", >>! dev+'a',hd_info[dev].cyl*hd_info[dev].head*hd_info[dev].sect/2048,hd_info[dev].cyl,hd_info[dev].head,hd_info[dev].sect); >> else { ... >>Once I sent this to Linus, but no response... > >You wouldn't believe how much stir-up a small patch like this can sometimes >cause. I think Linus has become more careful before incorporating things >like this.... Yeah. The above patch would cause linux to "recognize" non-existant devices that are entered in the BIOS tables but not physically present. -- mlord@bnr.ca Mark Lord BNR Ottawa,Canada 613-763-7482 ------------------------------ From: hugh@hugh.cosc.canterbury.ac.nz (Hugh Emberson) Subject: Re: Multiprocessing Pentium Systems Date: 13 Sep 1994 00:15:05 GMT >>>>> "Scott" == Scott Lawrence Lynn writes: Scott> In article , Scott> Hugh Emberson wrote: Scott> I've never looked at the SunOS 4.x.x kernel, but I can't Scott> imagine that it was done this way. Spinlocks have timeouts on Scott> them, and you could easily have a CPU wait for much too long Scott> due to the inherent possibility of starvation that comes with a Scott> spinlock. I hesitate to disagree with someone from Pyramid on something like this ... but I just spoke with our local SunOS expert and he says that there is a single mutex around the entire kernel in 4.1.3. You do get some serious resource contention problems. If you run top on a multiprocessor Sun running 4.1.3 you get something like this: Cpu states: 7.3% user, 0.0% nice, 7.4% system, 84.5% idle, 0.8% spin Right now, there isn't much happening, but when you get several processes which want to use lots of kernel time then the spin field gets upto 15% or higher. The spin field seems to measure the total time spent waiting to get into the kernel. Scott> One way to handle SMP simply is to put spinlocks around all the Scott> kernel data structures, or major subsystems. This will still Scott> probably take a great deal of work to get it right, and it'll Scott> have problems. It's a good start though. This is effectively what they have done. They consider the kernel to be one big data structure or subsystem :-) Of course the big problem with this is what to do with interrupts and I guess this will depend on the architecture. Which processor gets the interrupts? Can you designate which processor gets them, or do they only go to processor 0? If they only go to processor 0 then only processor 0 should be allowed to enter the kernel, cause we don't want to have a processor waiting to get into the kernel to handle an interrupt. And there are nasty coherence problems with the cache and tlb, but hopefully the hardware will handle some/most of these. I'm not suggesting this as a long term, elegant solution, but as a (hopefully) quick and dirty hack to get Linux running on these multiprocessor machines. Cheers, Hugh -- Hugh Emberson | ... from the end of the Information hugh@cosc.canterbury.ac.nz | Super-four-wheel-drive-track. ------------------------------ From: alan@spri.levels.unisa.edu.au (Alan Modra) Subject: Re: 3c509 Problems Date: 12 Sep 1994 07:41:10 GMT Danek Duvall (duvall@sage.wlu.edu) wrote: : I recently set up my 3Com Etherlink III Combo on my linux machine. : The first boot after the network stuff was configured, it worked fine. : In fact, it worked fine continuously for over a day. Then, today, I : was having some problems compiling and installing sendmail, I rebooted : my machine. At that point, I couldn't find anything on the network. : I hadn't changed any relevant pieces of the network config files, so : it couldn't have been that. Then I checked /var/adm/messages, which : had the line: : eth0: Missed interrupt, status then 2011 now 2011 Tx 00 Rx 383c : The same line appeared every time I booted, exced that the last number : would change. I found the spot where this gets printk'ed, but I know : nothing more than that. : Please! If there's anything anyone can do to help, I would really : appreciate it, since I have almost no clue about the networking code : (except what I've read in the NAG and the NET-2-HOWTO). If there's : any further information you need to figure out what's going on, please : get in touch. (My kernel version, by the way, is 1.1.49.) This one seems to be a 3c509 driver bug. I mailed Donald Becker a report way back in Dec about it, but it still isn't fixed (can't complain, I didn't fix it either :-) ). The most annoying result of the bug, for me, is that crond doesn't get uid's via NIS reliably, and sometimes your first login fails. My work-around is to put the following in rc.inet2 after ypbind is run. while ! /usr/bin/ypcat passwd >/dev/null; do /usr/bin/sleep 1; done ------------------------------ From: bjkramer@pluto.njcc.com (Brian Kramer) Crossposted-To: comp.os.linux.help Subject: Re: 3c509 Problems Date: 12 Sep 1994 19:17:34 -0400 : eth0: Missed interrupt, status then 2011 now 2011 Tx 00 Rx 383c Instead of rebooting turn off the machine (with shutdown of course) and leave it off for about a minute. I have the same card and if I do not do that you can't use the ethernet at all. I guess the card needs to reset. -- Brian Kramer - Owner/Systems Administrator - bjkramer@pluto.njcc.com New Jersey Computer Connection - Public Access Unix Site - pluto.njcc.com Voice: 609-896-2799 - Fax: 609-896-2994 - Dialups: 609-896-3191 Dialup or Telnet to pluto.njcc.com and log in as guest for more information. ------------------------------ From: ptjmatti@cc.Helsinki.FI (Petri T J Mattila) Subject: Interested in EIDE driver ? Date: 13 Sep 1994 08:52:02 GMT Hello world, I am working on eide driver for linux, and have the skeleton up and running. BUT, I don't have any information about eide controllers, (programming, technical, etc.) So, YOU could help me by mailing me info about - some site to ftp - some notes - anything concerning eide controllers I am also interested in what controllers are there, in general ?? SOME INFO ABOUT THE SKELETON DRIVER: - currently running on normal AT -bus (8 MHz) controller - speedup 10-15% compared to the previous code (hd.c with multimode) - two controller supported by default - drives them simultaneously - have lba addressing mode, autodetection, multimode, new pio modes, smart error recovery and many more - a test: 1.8 Mb/sec read, 1.7 Mb/sec write with Seagate decathlon in AT -bus (8MHz) So, If you have any information of controllers, feel free to mail it to me, I would appreciate it. I will release the code as soon as I get some timing bugs fixed. So, stay on line. Petri. -- ============================================================================= * Petri Mattila * ptjmatti@kruuna.helsinki.fi * University of Helsinki * ============================================================================= ------------------------------ Crossposted-To: comp.lang.fortran From: robertl@arnet.com (Robert Lipe) Subject: Compiler benchmarking. was Survey: who wants f77,cc,c++,hpf for linux? Date: Mon, 12 Sep 1994 18:11:12 GMT rsanders@mindspring.com (Robert Sanders) writes: >On Sun, 11 Sep 1994 17:31:04 GMT, mcastle@umr.edu (Mike Castle) said: >> In article , Dan Pop wrote: >>> Could you post some examples where a commercial native compiler for x86 >>> produces _significantly_ faster codes than the free gcc? >> IBM's C compilers under OS/2. >> Watcom compilers under OS/2 and DOS (do they have unix versions?) >> Zortech C compiler (I presume under OS/2 and DOS as well). >> Most likely the Mark Williams C compiler (they produce better >> compilers than operating systems (coherent)). >I think Dan meant he wanted to see the C source, timings, and the >assembly produced by each, as well as other relevant benchmark >information (input data, RSS, host machine, etc.). I'm certainly not trying to slam GCC - there's certainly a permanent place for it in my toolbox, but here is an excerpt from some internal tests I ran with identical dhrystone tests this spring....... All systems running SCO 3.2v4.2 or ODT 3.0. (Eliminating the OS vs. Executable part of the debate...) Amsg 25Mhz 486. One of the original Intel motherboards. Bytebox 486DX2/66. Generic Clone. Pooter 60Mhz Pentium Dell Notes: GCC is compiled with 486, so -m486 makes no diff. ICC is the Intel compiler (a.k.a. the "SCO Optimizing C Compiler") 1.3.11 RCC and CC are straight of the ODT 3.0 DS. Register was set in all cases. (Even though I believe the compiler should figure this stuff out, not me...) Set ts to 8 to read this. The same array of binaries were run on all three machines. While I'd like to convince you that I'd automated the hell out of this, I didnt - a lot of this is massaged by hand. I was just trying to decide if the compiler really makes any difference...I didn't want to make a career of it. First, the tallies by machine: 14039 amsg cc SCO ODT 3.0 compiler (ms 6.0 derived) 15639 amsg rcc -O The restricted c compiler (think it's pcc in 16056 amsg cc -O 22411 amsg gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 22416 amsg gcc -O2 -m486 -funroll-loop 22983 amsg cc -Ox 31113 amsg icc -O 31142 amsg icc Intel compiler. 41858 amsg icc -O -mem -ip 41893 amsg icc -O -mem -ip -p4 41963 amsg icc -O -mem -ip -p5 34211 bytebox cc SCO ODT 3.0 compiler (ms 6.0 derived) 39339 bytebox rcc -O The restricted c compiler (think it's pcc in 42408 bytebox cc -O 54525 bytebox gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 54525 bytebox gcc -O2 -m486 -funroll-loop 56274 bytebox cc -Ox 70372 bytebox icc -O 71581 bytebox icc Intel compiler. 95147 bytebox icc -O -mem -ip -p4 95877 bytebox icc -O -mem -ip 97560 bytebox icc -O -mem -ip -p5 50505 pooter cc SCO ODT 3.0 compiler (ms 6.0 derived) 56882 pooter cc -O 58445 pooter rcc -O The restricted c compiler (think it's pcc in 76887 pooter cc -Ox 86655 pooter gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 86730 pooter gcc -O2 -m486 -funroll-loop 139860 pooter icc -O 140252 pooter icc Intel compiler. 207039 pooter icc -O -mem -ip -p5 208333 pooter icc -O -mem -ip 208333 pooter icc -O -mem -ip -p4 Here's the same table, pumpted through reverse numeric sort. 208333 pooter icc -O -mem -ip -p4 208333 pooter icc -O -mem -ip 207039 pooter icc -O -mem -ip -p5 140252 pooter icc Intel compiler. 139860 pooter icc -O 97560 bytebox icc -O -mem -ip -p5 95877 bytebox icc -O -mem -ip 95147 bytebox icc -O -mem -ip -p4 86730 pooter gcc -O2 -m486 -funroll-loop 86655 pooter gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 76887 pooter cc -Ox 71581 bytebox icc Intel compiler. 70372 bytebox icc -O 58445 pooter rcc -O The restricted c compiler (think it's pcc in 56882 pooter cc -O 56274 bytebox cc -Ox 54525 bytebox gcc -O2 -m486 -funroll-loop 54525 bytebox gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 50505 pooter cc SCO ODT 3.0 compiler (ms 6.0 derived) 42408 bytebox cc -O 41963 amsg icc -O -mem -ip -p5 41893 amsg icc -O -mem -ip -p4 41858 amsg icc -O -mem -ip 39339 bytebox rcc -O The restricted c compiler (think it's pcc in 34211 bytebox cc SCO ODT 3.0 compiler (ms 6.0 derived) 31142 amsg icc Intel compiler. 31113 amsg icc -O 22983 amsg cc -Ox 22416 amsg gcc -O2 -m486 -funroll-loop 22411 amsg gcc -O2 -funroll-loop 2.5.6 i486-unknown-sco3.2v4.2 16056 amsg cc -O 15639 amsg rcc -O The restricted c compiler (think it's pcc in 14039 amsg cc SCO ODT 3.0 compiler (ms 6.0 derived) While I doubt this reflects anything like "real world" cases, one thing is evident: This Intel C compiler makes *HOT* code. Please, no flamage about how bogus the tests are, etc. I think my conclusion is correct - icc smokes gcc. I've stared out the assembly output of this beasty and it's incredible. Lessee, if ibcs is running correctly, an industrious linuxer could probably generate a version of gas that would read icc's output. This compiler is avilable from SCO and a few other vendors. ---- Robert Lipe Sr. Software Engr. Arnet Corp. robertl@arnet.com ------------------------------ Subject: Re: Alpha Linux From: boekhold@morra.et.tudelft.nl (Maarten Boekhold (Who'd you expect??)) Date: 13 Sep 94 11:25:46 +0200 N J Plant (nick@lepton.demon.co.uk) wrote: > In article > im14u2c@cegt201.bradley.edu "Joe Zbiciak" writes: > > > >In jem@snakemail.hut.fi (Johan Myreen) > >writes: > > > >>What is the natural word size of the 68000? Or the 8088? Or a > > > >Even better: The 68008... 8 bit data path, 16 bit registers, 32 bit ALU. > >By 32 bit ALU, I mean two registers would combine together and make a 32 bit > >register for ADD & SUB and MUL & DIV (I think.) > > > On the 68000 the external address bus is 20 bits and the external data bus > is 8. Internally, the registers, buses and ALU are all 32 bit. It can ADD > and SUBtract 32 bit numbers or MULtiply 2 16 bit numbers to give a 32 bit > result. It has fewer pins than a 68040, but its still a 32 bit chip. The > sizeof the integral types should be the same as any other 68K chip. Ho, hold it right there...... The 68000 has a 16 bit external databus. You're right about the other things though... :-) Maarten boekhold@morra.et.tudelft.nl ------------------------------ Subject: Re: Alpha Linux From: boekhold@morra.et.tudelft.nl (Maarten Boekhold (Who'd you expect??)) Date: 13 Sep 94 11:32:07 +0200 Chris Bitmead (chrisb@wombat.cssc-syd.tansu.com.au) wrote: > In article dholland@husc7.harvard.edu (David Holland) writes: > >adc@bach.coe.neu.edu's message of 06 Sep 1994 16:38:15 GMT said: > > > > > Why drop one? > > > 16 bits = short int > > > 32 bits = int > > > 64 bits = long > > > >Over in the next thread people were talking about Unicode; why not > > > >16 bits = char > >32 bits = short > >64 bits = int, long > None of these is the best solution. The best solution is to say exactly > what you mean. E.g. If you want to store numbers between -500 and +1000 > you should declare this and let the compiler work out how many bits to > use. e.g. int{-500,1000} foo; int{0,65535} bar; > Naturally you could use typedef's for common ranges. This has the added > benefit of not assuming that the computer is binary, assuming one day > someone invents a non-binary computer. > In any case stating what you *really* want to do and letting the compiler > decide on the optimum number of bits has to be the best. I really like this solution. It might take some time to get used to, but I like the idea. Maarten boekhold@morra.et.tudeft.nl ------------------------------ From: schrod@iti.informatik.th-darmstadt.de (Joachim Schrod) Subject: Re: Acid Date: 13 Sep 1994 10:43:29 GMT In article , turnbull@turnbull.sk.tsukuba.ac.jp (Stephen J. Turnbull) writes: > > > Actually, TeX/LaTeX would come closest...and its available for several > >languages. > > Including Chinese and Japanese. But I don't know if Don Knuth thought > about multidirectional languages. There is support for typesetting approximately 50 languages, including such beasts as Khmer. Multi-directional, of course. There exists also a Unicode-TeX. Don't expect the stuff to be easy to install, though. It still have its rought edges, in particular on the font side. But I did see it work... :-) Joachim -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Joachim Schrod Email: schrod@iti.informatik.th-darmstadt.de TeX Users Group Technical Working Group on Multilingual Coordination, member ------------------------------ Crossposted-To: comp.lang.fortran From: iialan@iifeak.swan.ac.uk (Alan Cox) Subject: Re: Survey: who wants f77,cc,c++,hpf for linux? Date: Tue, 13 Sep 1994 09:53:00 GMT In article <1994Sep11.173104.10758@umr.edu> mcastle@umr.edu (Mike Castle) writes: >In article , Dan Pop wrote: >>Could you post some examples where a commercial native compiler for x86 >>produces _significantly_ faster codes than the free gcc? >Watcom compilers under OS/2 and DOS (do they have unix versions?) I've benchmarked the Watcom compiler (pre-pentium one which they claim is much quicker) and gcc beat it fairly soundly. >Zortech C compiler (I presume under OS/2 and DOS as well). You must be kidding. >Most likely the Mark Williams C compiler (they produce better > compilers than operating systems (coherent)). It produces code about as fast or fractionally slower but much much smaller, and although its not a proper ANSI compiler so speed comparisons are not totally fair it compiles much quicker than gcc. >GCC is designed to be PORTABLE first, optimal last. In almost >all cases, a DECENT architecture specific compiler will be as >good as or beat GCC. Very few compilers do beat it - Sun's sparc compiler does. The DOS compilers I would like to have a good benchmark for but they are 16bit currently. Some figures for 32 bit Borland code would be interesting because the 16bit code it writes is very very good. Alan -- ..-----------,,----------------------------,,----------------------------,, // Alan Cox // iialan@www.linux.org.uk // GW4PTS@GB7SWN.#45.GBR.EU // ``----------'`----------------------------'`----------------------------'' ------------------------------ ** FOR YOUR REFERENCE ** The service address, to which questions about the list itself and requests to be added to or deleted from it should be directed, is: Internet: Linux-Development-Request@NEWS-DIGESTS.MIT.EDU You can send mail to the entire list (and comp.os.linux.development) via: Internet: Linux-Development@NEWS-DIGESTS.MIT.EDU Linux may be obtained via one of these FTP sites: nic.funet.fi pub/OS/Linux tsx-11.mit.edu pub/linux sunsite.unc.edu pub/Linux End of Linux-Development Digest ****************************** .