From: Digestifier To: Linux-Development@senator-bedfellow.mit.edu Reply-To: Linux-Development@senator-bedfellow.mit.edu Date: Fri, 14 Oct 94 13:13:15 EDT Subject: Linux-Development Digest #307 Linux-Development Digest #307, Volume #2 Fri, 14 Oct 94 13:13:15 EDT Contents: Re: ext2fs vs. Berkeley FFS ("Theodore Ts'o") Re: Shared Libs: working toward a permanent solution? ("Theodore Ts'o") accept() not functioning 100%? (Robert Simons) Re: Improving SLIP latency under Linux (Frank Lofaro) Re: Status of Linux and Distributions security (Bernd Eckenfels) [Q] touch-thru-glass keypad driver suggestions (Joe Desbonnet) Re: Suggestion: comp.os.linux.channelecho.* (Matthias Urlichs) A/D-board with Linux-driver?? (Sebastian Egner) Re: libg++-2.6: builtinbuf undefined (David Fox) Re: Shared Libs: working toward a permanent solution? (Bryan Ford) Re: A badly missed feature in gcc (Dan Pop) Re: Status of Mac Linux & PPC Linux? (Thomas G. McWilliams) ypbind didn't work (Ari Widodo) 1.1.52 Kernel Panics (Christopher Wiles) .52 & .52 Hangs in heavy FTP'ing.. (Thomas Lundquist) Re: LINUX Logical volumes (Werner Almesberger) Re: 3c505 driver ? (John Huang) Term blocks modem, switching to VT and back restores? (Rafal Kustra) ---------------------------------------------------------------------------- From: "Theodore Ts'o" Subject: Re: ext2fs vs. Berkeley FFS Date: 14 Oct 1994 09:17:48 -0400 Reply-To: tytso@MIT.EDU From: mike@majestix.cs.uoregon.edu (Mike Haertel) Date: 03 Oct 1994 20:46:49 GMT I'd rather have garbage in a file than have a file whose blocks are also in the free list and will soon be allocated to another file as well. Of course, fsck is supposed to fix these situations. But if you're using a filesystem wherein the actual disk updates are trailing way behind file system activity, and are not constrained to be written in the order they were made, then a crash at a disk-intensive time could leave fsck extremely confused. This is why BSD ffs does its updates in what somebody called the "wrong" order, but which strikes me as exactly the _right_ order. ext2 doesn't use a free list, but rather uses a bitmap of used/free blocks. Making sure this bitmap is correct is perhaps the easiest job e2fsck has to do. That's *not* what's going to get it confused. The one place where a lagged write of the meta data might get the e2fsck confused is if indirect block of a deleted file gets overwritten by something else, probably another data block, before the inode table gets flushed to disk. In this case, e2fsck *will* notice the inconsistency; however, it will probably call for human intervention in order to fix the problem. Of course, I've tried to make e2fsck such that the human intervention is as painless as possible. For example, e2fsck can automatically take care of cloning cross-linked files so that each file has its own copy of blocks claimed by two different inodes. Of course, at least one of those files will have garbaged blocks (consisting of data from the other file) after this operation, but at least the filesystem is stable. The human being simply has to examine both of the files and see which one should be discarded. In any case, it's a good idea to optimize for the common case, not the exception case. Thus, I believe ext2fs made a pretty good tradeoff. Not doing syncronous meta-data writes gives the ext2fs a blinding speed advantage over the FFS. From my experience with both FFS and ext2fs, ext2fs is just as reliable --- if not more --- than FFS in surviving unclean shutdowns. - Ted ------------------------------ From: "Theodore Ts'o" Subject: Re: Shared Libs: working toward a permanent solution? Date: 14 Oct 1994 09:22:09 -0400 Reply-To: tytso@MIT.EDU The real problem with this sort of solution is that you have to educate GCC about segment registers, and you have to include the segment number into a pointer. Otherwise, how is GCC supposed to know whether a pointer was pointing at memory from the library's data segment, or at memory passed in from the caller? I've discussed this problem with GCC hackers who were also Multics weanies, and this is a very difficult problem. If it could be solved, then we could indeed put shared libraries (and even shared memory!) into separate segments, ala Multics. This would be very clean from an OS architectural issue --- unfortunately, it requires a lot of compiler support that just doesn't exist right now. It would also break a lot of programs, since a pointer would be 48 bits, instead of 32 bits, and there are a lot of programs that assume that the width of pointer is the same as the width of an integer, or a long. (This isn't a portable assumption --- but there are still a lot of programs left over from the "All the world's a Vax" days.) - Ted ------------------------------ From: Robert_Simons@sedl.com (Robert Simons) Subject: accept() not functioning 100%? Date: 12 Oct 94 07:03:04 +0500 EST According to every manual I have ever read, the accept() system call will load the passwed sockaddr structure with information about the connecting client. According to my tests, this doesn't happen under Linux. Consider: #define ADDRESS 3012 /* Some unused user-defined port */ dumb_func() { char c; FILE *fp; int fromlen; register int i, s, ns, len; struct sockaddr_un saun, fsaun; for(;;) { /* Set up signal handler for kill 15 and Interrupt */ signal(SIGTERM, cleanup); signal(SIGINT, cleanup); /* Get a socket to work with. The socket will be in the UNIX domain, and will be a stream socket */ if ((s = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) { perror("server: socket"); exit(1); } /* Create the address we will be binding to */ saun.sun_family = AF_UNIX; strcpy(saun.sun_path, ADDRESS); /* Try to bind the address to the socket. We unlink the name first, so that the bind won't fail. The third argument indicates the "length" of the structure, not just the length of the socket name! */ unlink(ADDRESS); len = sizeof(saun.sun_family) + strlen(saun.sun_path); /* fromlen = sizeof(fsaun.sun_family) + strlen(fsaun.sun_path); */ fromlen = sizeof(fsaun); if (bind(s, &saun, len) < 0) { perror("server: bind"); exit(1); } /* Listen on the socket */ if (listen(s, 5) < 0) { perror("server: listen"); exit(1); } printf("\nWaiting for connection: \n"); /* Accept connections. When we accept one, ns will be connected to the client. fsaun will contain the address of the client. */ if ((ns = accept(s, (struct sockaddr_un *)&fsaun, (int *)&fromlen)) < 0) { perror("server: accept"); exit(1); } printf("Connection from: %s\n", fsaun.sun_path); printf("Connection flen: %d\n", fromlen); /* We'll use stdio for reading the socket */ fp = fdopen(ns, "r"); /* We can simply use close() to terminate the connection, since we are done with both sides */ close(s); } exit(0); } In the above example, the fsaun instance of sockaddr_un is not getting loaded properly. Any ideas? Everything seems to work fine, it just doesn't load the client information into the sockaddr structure... --- rsimons@sedl.com [Computer Online Services (COS) - Clique.com] [3606 10th Ave North. Palm Springs, FL 33461] [Voice.............407-641-7454] [Modem.............407-641-7484] [FAX...............407-641-7226] ------------------------------ From: ftlofaro@unlv.edu (Frank Lofaro) Subject: Re: Improving SLIP latency under Linux Date: Fri, 14 Oct 94 02:51:57 GMT In article <37981n$9bs@gate.noris.de> urlichs@smurf.noris.de (Matthias Urlichs) writes: >Oops, I meant "-2- or -3-". (Rationale: There should be a link-level >checksum. The modem error correction isn't end-to-end, but it's better than >no error correction at all.) > Huh??? You need link-level checksums when you already have TCP and UDP checksums? ------------------------------ From: ukd1@rzstud1.rz.uni-karlsruhe.de (Bernd Eckenfels) Subject: Re: Status of Linux and Distributions security Date: 14 Oct 1994 02:40:54 GMT Cees de Groot (cg@tricbbs.fn.sub.org) wrote: : In article <36o1rs$8bh@vixen.cso.uiuc.edu>, : Daniel L. Marks wrote: : >How do the recent linux kernels (1.1.45+) and the various distributions : >(Slackware, SLS, Debian) compare to commercial UNIX offerings such as : >SCO, UnixWare, and BSDI for the number of security holes each is known : >to have? There is a Problem with the memory management. If u request huge amounts of memory without writing to it, u can request more memory than virtually are available. If u now start to write into those pages, the free memory will decrease, the system will swap more and more, until it is really locked up. Something like SIGWARN (from AIX) should be impemented, or at least some appropirate (on user basis) rlimits. Greetings Bernd PS: i patched a quick and drity RLIMIT_NPROC without hte use of an structure for user-informations. I think it is time to invent an structure wich is allocated once for each user (a process is running for), wich can be used to restrict the maximu numbers of processes for an user, or the maximum amount of virtual memory. This is not possible on an process-basis. -- (OO) -- Bernd_Eckenfels@Wittumstrasse13.76646Bruchsal.de -- ( .. ) +4972573817 ecki@lina.ka.sub.org ukd1@rz.uni-karlsruhe.de o--o *QUAK* Jetzt auch mit Plueschtier in der .Sig! (O____O) Eckes@IRC ------------------------------ From: joe@iol.ie (Joe Desbonnet) Subject: [Q] touch-thru-glass keypad driver suggestions Date: 12 Oct 1994 08:50:29 GMT I'm writing a driver for a touch-thru-glass keypad. The unit sends codes via a serial line and I have successfully written code to decode this. However I need suggestions as to how I can effectivly press a key on the keyboard so that this will work with applications without modifying them. I guess it should be possible to program this in user mode(?). Thanks, Joe Desbonnet. ------------------------------ From: urlichs@smurf.noris.de (Matthias Urlichs) Subject: Re: Suggestion: comp.os.linux.channelecho.* Date: 12 Oct 1994 08:37:46 +0100 In comp.os.linux.development, article , hpa@nwu.edu (H. Peter Anvin) writes: > > Forge a message using > /usr/lib/sendmail -f from-address linux-activists-request@niksula.hut.fi > > Note: requires root access, and you to type in all relevant headers. > I know that. Tried it, didn't work. (Not even a peep out of the list processor.) The next step was to telnet to their SMTP port. Still no luck. And, yes, I know RFC822; the message _was_ correct. -- The only winner of the War of 1812 was Tchaikovsky. -- Solomon Short -- Matthias Urlichs \ XLink-POP Nürnberg | EMail: urlichs@smurf.noris.de Schleiermacherstraße 12 \ Unix+Linux+Mac | Phone: ...please use email. 90491 Nürnberg (Germany) \ Consulting+Networking+Programming+etc'ing 42 PGP: 1B 89 E2 1C 43 EA 80 44 15 D2 29 CF C6 C7 E0 DE Click here. ------------------------------ From: egner@hermes.zkm.de (Sebastian Egner) Subject: A/D-board with Linux-driver?? Date: 14 Oct 1994 14:05:49 GMT Does anybody know of multiple channel analog/digital interface boards for AT-bus IBM-compatibles for which there is regular (/dev/*) Linux device driver software? Sebastian Egner, Zentrum f"ur Kunst und Medientechnologie, Karlsruhe (Germany) ------------------------------ From: fox@graphics.cs.nyu.edu (David Fox) Subject: Re: libg++-2.6: builtinbuf undefined Date: 14 Oct 1994 02:52:37 GMT In article sattler@iti.cs.tu-magdeburg.de (Kai-Uwe Sattler) writes: ] > Error message: ] > ../../libg++.a(stdstrbufs.o): Undefined symbol builtinbuf referenced from data segment ] ] I have the same problem, but only with Linux. On SunOS 4.1.3 and ] Solaris 2.3 I can build libg++ without errors. Any hints ? I got the error on SunOS 4.1.3 too. I just removed stdstrbufs.o from the list of files to be made, I haven't missed it yet. -- David Fox xoF divaD NYU Media Research Lab baL hcraeseR aideM UYN ------------------------------ From: Bryan Ford Subject: Re: Shared Libs: working toward a permanent solution? Date: 14 Oct 1994 11:12:22 -0400 Reply-To: baford@schirf.cs.utah.edu |The real problem with this sort of solution is that you have to educate |GCC about segment registers, and you have to include the segment number |into a pointer. Otherwise, how is GCC supposed to know whether a |pointer was pointing at memory from the library's data segment, or at |memory passed in from the caller? It doesn't need to. Any pointer that actually gets passed around in a program in variables and such is still a 32-bit, absolute, "small model" pointer as usual. The only thing that needs to be changed is the method of accessing or taking the address of a static variable. To take the address of a static variable, just convert the segment-register-relative offset to a global, absolute offset by adding the base address of the library, and use the resulting absolute pointer thereafter. This will work as long as a particular shared library doesn't move around in a particular process's address space _while_it_is_running_, which would be rather insane anyway. :-) |I've discussed this problem with GCC hackers who were also Multics |weanies, and this is a very difficult problem. If it could be solved, |then we could indeed put shared libraries (and even shared memory!) into |separate segments, ala Multics. This would be very clean from an OS |architectural issue --- unfortunately, it requires a lot of compiler |support that just doesn't exist right now. Yes, a true multisegmented environment has its attractions, but after programming in one for a while (i.e. 16-bit DOS), I'd say its disadvantages are much bigger than its advantages. :-) But I wasn't proposing a true multisegmented environment; just a small architecture-specific kludge to allow the compiler to use a segment register instead of a scarce general register to remember the library base address in a PIC library. Bryan ------------------------------ From: danpop@cernapo.cern.ch (Dan Pop) Subject: Re: A badly missed feature in gcc Date: Thu, 13 Oct 1994 11:10:10 GMT In <37i5mu$ba4@harbinger.cc.monash.edu.au> kevinl@fangorn.cs.monash.edu.au (Kevin Lentin) writes: >So are comments. The only non-C things the compiler sees are the # nn file >Directives so it knows where source comes from. So there is still no >meaning to the carriage return added to the comment syntax as far as the >compiler is concerned. > >That aside, I believe C++ comments are better. They're great for single >line comments, make temporary removal of a line easy (I still use C >comments for temporary removal of small blocks of code and #if for bigger >ones). BUT I think C should remain C. You're just going to break something >somewhere eventually. Besides breaking valid C code, changing cpp to accept // comments as default behaviour will also break other utilities which rely on cpp. For example, if you have in your .Xdefaults: Mosaic*homeDocument: http://info.cern.ch the result will be completely different than what you expect, unless you use the -nocpp xrdb option, which means removing functionality from xrdb. Another example is the Unix Fortran compiler, f77. If it compiles a file with the .F extension, it will pass the source file through cpp before compiling it. But // is a Fortran operator! :-) Dan -- Dan Pop CERN, CN Division Email: danpop@cernapo.cern.ch Mail: CERN - PPE, Bat. 31 R-004, CH-1211 Geneve 23, Switzerland ------------------------------ Crossposted-To: comp.os.linux.help From: tgm@netcom.com (Thomas G. McWilliams) Subject: Re: Status of Mac Linux & PPC Linux? Date: Wed, 12 Oct 1994 08:16:33 GMT The following was posted by gary@ah3.cal.msu.edu (Gary J LaPointe) to comp.sys.powerpc: ======================================================================= LINUX ON POWER MAC DELAYED AS I/O PROBLEMS & COLLEGE WORK INTERVENE (October 7th 1994) The amateur development work to port the Linux version of Unix to the Power Macintosh has stalled. Back in May (Issue 5 story 1054), two students, Charlton Wilbur and Jem Lewis, laid out their ambitious plans to get the Freeware Unix clone up and running on Apple's machines by September. Lewis admits that finding spare time is a problem. However he said this week: "Our biggest problem besides time right now is that the Power Mac I/O stuff is a proprietary technology, and Apple's not telling." He adds that they are trying to "cajole" the necessary information from Apple and are also looking at how to reverse engineer the chip. Lewis says that the pair are still plodding on with the project; "When we think we have a handle on what's going to happen with the Mac I/O stuff, we'll send a note with a revised schedule", he said. -- ------------------------------ From: ari@sinec.cent.saitama-u.ac.jp (Ari Widodo) Subject: ypbind didn't work Date: Wed, 12 Oct 1994 10:26:49 GMT First, I would like to say thanks for all whom gave me information about NIS. And I got the yp-client packages and install that. But I canN't run ypbind. --- /usr/sbin/ypbind Undefined C library functions: 1. light C shared image (Use the real one instead.) --- Should I recompile my C library ? Can somebody give me some explanation for this error ? Thank you very much, Regards, Ari Widodo -- ===== A R I W I D O D O E=mail: ari@cent.saitama=u.ac.jp ===== Saitama University Dept. of Electrical and Electonic Engineering Member of ACCESS, Information Processing Center. -- ===== A R I W I D O D O E=mail: ari@cent.saitama=u.ac.jp ===== Saitama University Dept. of Electrical and Electonic Engineering Member of ACCESS, Information Processing Center. ------------------------------ From: a0017097@wsuaix.csc.wsu.edu (Christopher Wiles) Subject: 1.1.52 Kernel Panics Date: Mon, 10 Oct 1994 19:03:57 GMT This has happened twice in the past two days: The machine has been running for a few hours. I'm using EMACS on a short (<5Kb) file. Suddenly, I get a kernel panic along the lines of "EXT2FS: can't access inode" (I apologize for not having the exact error message, I can't find the scrap of paper I wrote it on). Three-finger-salute does not work, machine locks up nicely, and I have to press the reset button. The machine in question is a 386DX40, 8 megs of RAM, two Maxtors (a 42 meg and a 540 meg). The kernel was compiled the day before yesterday with GCC 2.6.0, -O2, -m386. I'm fairly sure that it isn't GCC, as I successfully used a 2.6.0-compiled 1.1.51 for almost a month, no incident. As I've got a really nasty software engineering project due Thursday, I do not have the time to debug this one. I've went back to 1.1.51 (umount bugs and all). Has anyone else experienced this phenomenon? -- Chris a0017097@wsuaix.csc.wsu.edu wileyc@halcyon.com wileyc@quark.chs.wa.com "... but I want to use all eight comm ports SIMULTANEOUSLY!" PGP 2.6 public key available by finger for the clinically paranoid. ------------------------------ From: thomasez@dhhalden.no (Thomas Lundquist) Subject: .52 & .52 Hangs in heavy FTP'ing.. Date: Thu, 13 Oct 1994 17:03:17 GMT Hi. I have problems running, 1.1.53 in my box. It will hang with no notice when people is ftp'ing to it. They use different DOS/WIN ftp's. It seems like 1.1.51 works, but I have encountered (sp!) hangs there too.. but more casually. I have used the .tar file from funet.fi with 1.1.52. Setup: 486dx66 16Mram 800M IDE HD (yes, I should by a scsi drive..:=) 2.1G, 325M ScsiHD aha 1542B generic S3 card Ne2000 (original) GUS w/1MB (tried both with sound in kernal and without.) Guess this is a betareport.. WIsh I had more info, but nothing in /usr/adm/ gives me any hints. Thanks. -- Thomas Lundquist. eMail : ThomasEZ@hiof.no Oestfold regional college, Norway. Baldrick @Final Realms mud Departement of computer science. @kark.hiof.no port 2001 Kark = 158.36.33.5 ..PGP key available.. ..Peace.. ------------------------------ From: almesber@nessie.cs.id.ethz.ch (Werner Almesberger) Subject: Re: LINUX Logical volumes Date: 13 Oct 1994 10:57:03 GMT In article marcus@ee.pdx.edu writes: >In-reply-to: adc@bach.coe.neu.edu's message of 12 Oct 1994 15:34:56 GMT > Logical volumes should be easy to implement with userfs. But then they'd be slow. I think there are three distinct areas that are typically associated with the concept of logical volumes: 1) flexible disk space assignment: a virtual storage pool 2) higher data troughput: striping accross multiple disks 3) more reliability: checksum drives and/or mirroring 2) and 3) are typically associated with RAID concepts. 1) would be fairly simple to implement, with only a marginal performance hit. 2) could be easy too, but it may be hairy to administer (e.g. you'd have to make good guesses about the optimal length of physical reads. Of course, that may change from cylinder to cylinder.) Another problem with 2) is, that many people would want to be able to use it with sets of non-identical drives, which would give only partial speed improvements and/or increase the complexity of configuring such a system. 3) is the hardest part. For 3), you need some extended disk access scheduling, e.g. write data, THEN write copy (or update checksum), THEN allow any other writes to related sectors/blocks. Also, some cheap mechanism to keep the system state after a failure (e.g. power drops and system is interrupred in the middle of a write, so your block may be partially corrupted but you can't flag is as "bad" anymore) would be mandatory. Of course, write buffering in the drive could make all this just impossible. Also, 3) requires some good understanding of failure patterns. Since nowadays, the electronic parts of a disk drive tend to be more likely to fail that the mechanic parts, you also have to be concerned about disk controller or (very common !) bus problems. Of course, an easy but probably adequate solution to the latter would be to declare them as something that's beyond the scope of such a concept. Let's also not talk about drive firmware bugs (and yes, you see more of those in large, less common setups). Since 1) and 2) significantly increase the average amount of data lost per time period (if any single drive of a large virtual drive fails, you might lose an entire huge file system, so the loss increase can be even quadratic), you either want to have a very good backup policy or, some security of the 3) type. Because people generally tend to neglect backups, the result of implementing 1) and/or 2) without 3) might be that the general impression of logical volumes would be that they don't work well and that Linux itself might be unreliable. (Such generalizations happen all the time - just witness the many claims of file system errors even if it's very obvious that hardware problems are the real cause.) Another problem is that building 1) and 2) without taking into account (i.e. prototyping) 3) might result in a design that makes it hard to add 3) at a later time. So, as a poster has already commented, it's not the implementation, it's the design. > What about `ifs'? IFS can be used to extend your file system capacity, but you pay a very high price in terms of access latency and administrative overhead if you use it this way, even if you take the old kernel-based one. (And then you also get a bonus share of unreliability.) - Werner -- _________________________________________________________________________ / Werner Almesberger, sending this from almesber@nessie.cs.id.ethz.ch / /______________________..._but_now_at_home_at________almesber@di.epfl.ch_/ ------------------------------ From: huang@ctt.com (John Huang) Subject: Re: 3c505 driver ? Date: 14 Oct 1994 16:29:15 GMT In article <1994Oct13.110404.20492@uxmail.ust.hk>, Tall Sword wrote: >Zheng Huang (huang@eagle.sangamon.edu) wrote: >: Hi, > >: I tried to config my linux(slakware 2.0) to work with my 3c505 network card, >: but I can't find the 3c505.c file in linux src directory under drivers. >: For some reason, the 3c505 is been commented out from the config file. Could >: someone tell me how I can make 3c505 work with linux. >Try get the update kernel and compile again. For me, I am running 1.1.53 and >the 3c503 works fine on my system. > >-- >* Origin: TallSword, Computer Science Year 2, HKUST > internet: cs_kokim@dmf123.ust.hk, cs_kokim@stu.ust.hk > root@dmf123.ust.hk, raymond@dmf123.ust.hk > Raymond.Ko@f15.n700.z6.ftn.air.org > fidonet: Raymond Ko, 6:700/15@fidonet.org > Does 3c503 driver works with 3c505 card? I tried to configure the kernel to include the 3c503 driver but it don't recognize my card. I know the linux has support for all the 3com card except for 505. I tried my 3c501 card, and it works. Any help for 3c505 driver? Thanks for the reply. -- John huang@image.ctt.com ------------------------------ Crossposted-To: comp.windows.x.i386unix From: rafal@cs.toronto.edu (Rafal Kustra) Subject: Term blocks modem, switching to VT and back restores? Date: 11 Oct 94 02:18:17 GMT I have ATI GUP, 2MVRAM. When I run term and do intensive communications (xmosaic over term, or some other graphical stwre), I would get my modem blocked. Term just keeps sending data from remote, with higher and higher latencies. Nothing will happen on my machine. In the past, I would have to kill term and seyon, and even than I wouldn't be able to connect to modem. Sometimes modem would send responses to requests made minute ago, as it if was backloged. But now I found a "solution": I such a blockage occurs (for example, I start xmosaic, having already opend emacs, all via term, and see that nothing is happening, I move my pointer to emacs window, and see that cursor would no highlight.) than I switch to VT and back to X, and some progress will be made. I just went over 20 or so pages of htlm in mosaic, having had to switch to VT and back few times. It (switching) just seems to unblock the modem. I am running Xfree 3.1, but same kind of blocking occurs in 2.1.1. I haven't tested whether this switching seems to restore the modem in 2.1.1, but I remember now it seem to did, concidentaly. It ("solution") just occured to me now. Thing that bothers me are the IRQ jumpers on my ATI/GUP board, with choice for IRQ3/2/5. However, manual says not to touch this and all are open now. I tried, some time ago, set it to IRQ 5, but I don't think it alleviated the problem - I remmember still having those loockups, although I was looking for smthg else then. (My modem is running on COM2/IRQ 3 with 14400B/56000 DTE speed. Also tried 38400 DTE). Also ATI is said to use I/O address normaly belonging to COM 4, and my I/O card is set at COM1/COM4 (since COM 2 is used by internal modem). I cannot disable COM2/COM4 alltogether. I reiterate: I do not have problems with Video card, and modem/term, as long as intensive (bi-directional, it seems) graphic related term communication takes place. For example, I usually do not have problems with uploading /downloading files, in term, even at the same time. Same for running EMACS19 over term. Xmosaic, is an almost sure way to block the term/modem couple. Any ideas? (I am running Linux 1.1.52, same happened since at least 1.1.37. Before it was rather OK, although I did not do much with term. Also I got my ATI/GUP at about the same time.). Thanx, Rafal ------------------------------ ** 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 ****************************** .