* Simple Linux/MIPS 0.2b @ 2000-07-13 7:16 Keith M Wesolowski 2000-07-13 13:58 ` J. Scott Kasten 0 siblings, 1 reply; 23+ messages in thread From: Keith M Wesolowski @ 2000-07-13 7:16 UTC (permalink / raw) To: linux-mips Simple Linux/MIPS 0.2b ``Death on a Stick'' has been released. *** This is not a production release *** This version runs on big-endian MIPS systems and has been tested on R4400 SGI IP22. You can obtain the distribution from the following location: MANDATORY Release Notes: ftp://oss.sgi.com/pub/linux/mips/mips-linux/simple/userland-0.2b/RELEASE_NOTES-0.2b Binaries: ftp://oss.sgi.com/pub/linux/mips/mips-linux/simple/userland-0.2b/bin There are also some patches in the src/ directory. I have lost some of my patches but will be recreating them and making them available at that location as well. Also included is a summary of the commands used to build the distribution. Highlights: - glibc 2.1.90 (000622) based - with empty symbol fix from Maciej - gcc 2.96 (000707) - binutils 2.10.90 (000707) - linux 2.4.0-test3-pre5 (000708) - support for profiling - full netkit and net-tools - gdb 5.0 - ncurses 5.1 - kernel debugging tools - numerous bugs fixed - numerous new bugs Please read the release notes before using. Installation instructions for 0.1, which should still work in general, can be found at http://foobazco.org/~wesolows/Install-HOWTO.html. Thanks to everyone who has been submitting patches, fixing bugs, writing the software, and in general making this distribution possible. This will be the last release for at least 6 weeks. Happy hacking! -- Keith M Wesolowski <wesolows@foobazco.org> http://foobazco.org/~wesolows/ (( Project Foobazco Coordinator and Network Administrator )) aiieeeeeeee! "The list of people so amazingly stupid they can't even tie their shoes?" "Yeah, you know, /etc/passwd." ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 7:16 Simple Linux/MIPS 0.2b Keith M Wesolowski @ 2000-07-13 13:58 ` J. Scott Kasten 2000-07-13 22:51 ` Ralf Baechle 2000-07-13 23:30 ` Keith M Wesolowski 0 siblings, 2 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-13 13:58 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips On Thu, 13 Jul 2000, Keith M Wesolowski wrote: > Simple Linux/MIPS 0.2b ``Death on a Stick'' has been released. Fantastic! I've had great success with release 0.1 on an Indy. Perhaps it's almost too stable for my blood. Need to go play with fire! ;-) Actually, I do have a question for any of you that have been doing builds. I seem to be having some interesting problems with shared object libraries. I'm using S-L-M 0.1 on an Indy as a development/test bed for code that will eventually end up in an embedded vrXXXX system. (I've already verified that the binaries are portable.) However, I am having difficulty on the Indy itself. NONE of the shared objects that I build on that box can be used by programs on that box. Yes, I either placed them in /lib or /usr/lib or built a proper ld.so.cache file for them. The ldd utility can resolve the libraries, etc... However, every program built on that box against libraries that I built on that box exits with an immediate bus error. It appears to be an issue with the libraries themselves. If I scavenge the equivalent libraries from the Hard Hat 5.1 distro and drop them in place, everything seems happy and works fine. Thus, there's something funky about the so's I'm building myself. I've tried building the XFree 4.0.1 libraries, Lesstiff libraries, and libraries from our own code base. All breaks the same way. #1 What form of black magic are you guys using to do this? #2 Are there specific compiler flags/phases that should/shouldn't be used with MIPS arch so's that are different than what I'd normally do under Linux? #3 Should I be trying to cross compile the so's them selves instead of native builds? #4 Are you using different flavors of gcc/binutils to do different jobs because of known breakages? Any tips appreciated here. In the mean, I'm going to start playing with flags. Sincerely, -JSK- ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 13:58 ` J. Scott Kasten @ 2000-07-13 22:51 ` Ralf Baechle 2000-07-13 22:39 ` J. Scott Kasten 2000-07-19 14:04 ` Maciej W. Rozycki 2000-07-13 23:30 ` Keith M Wesolowski 1 sibling, 2 replies; 23+ messages in thread From: Ralf Baechle @ 2000-07-13 22:51 UTC (permalink / raw) To: J. Scott Kasten; +Cc: Keith M Wesolowski, linux-mips On Thu, Jul 13, 2000 at 10:58:48AM -0300, J. Scott Kasten wrote: > distro and drop them in place, everything seems happy and works fine. > Thus, there's something funky about the so's I'm building myself. > > I've tried building the XFree 4.0.1 libraries, Lesstiff libraries, and > libraries from our own code base. All breaks the same way. > > #1 What form of black magic are you guys using to do this? Holy Penguin Pee (TM). Available at www.antarctica.com.br ;-) > #2 Are there specific compiler flags/phases that should/shouldn't be used > with MIPS arch so's that are different than what I'd normally do under > Linux? No compiler flags needed. > #3 Should I be trying to cross compile the so's them selves instead of > native builds? That should make no difference in the resulting binaries if you use the same versions of everything. > #4 Are you using different flavors of gcc/binutils to do different jobs > because of known breakages? We have various known problems with the various binutils version around. We're working on getting a current snapshot of binutils working properly but right now we still have various problems, therefore egcs 1.0.3a + binutils 2.8.1 is still the recommended version. Ralf ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 22:51 ` Ralf Baechle @ 2000-07-13 22:39 ` J. Scott Kasten 2000-07-19 14:04 ` Maciej W. Rozycki 1 sibling, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-13 22:39 UTC (permalink / raw) To: Ralf Baechle; +Cc: Keith M Wesolowski, linux-mips On Fri, 14 Jul 2000, Ralf Baechle wrote: > We have various known problems with the various binutils version around. > We're working on getting a current snapshot of binutils working > properly but right now we still have various problems, therefore > egcs 1.0.3a + binutils 2.8.1 is still the recommended version. OK, I'll give that a shot along with the binutils-000707, gcc-000707 on the Simple Linux site. I've disovered that SOME of the shared objects I build work, some do not at this point using the egcs-1.1.2 and binutils-2.9.5 from the first Simple release. I'll keep hammering at this until I find a combo that does the job. Thanks. ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 22:51 ` Ralf Baechle 2000-07-13 22:39 ` J. Scott Kasten @ 2000-07-19 14:04 ` Maciej W. Rozycki 2000-07-19 14:19 ` Hiroshi Kawashima ` (2 more replies) 1 sibling, 3 replies; 23+ messages in thread From: Maciej W. Rozycki @ 2000-07-19 14:04 UTC (permalink / raw) To: Ralf Baechle; +Cc: J. Scott Kasten, Keith M Wesolowski, linux-mips On Fri, 14 Jul 2000, Ralf Baechle wrote: > We have various known problems with the various binutils version around. > We're working on getting a current snapshot of binutils working > properly but right now we still have various problems, therefore > egcs 1.0.3a + binutils 2.8.1 is still the recommended version. I have binutils 2.10 and gcc 2.95.2 which appear to be stable, i.e. I haven't observed any problems recently. I may publish patches if anyone is interested. I may see if I can arrange to publish RPM packages as well. -- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available + ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:04 ` Maciej W. Rozycki @ 2000-07-19 14:19 ` Hiroshi Kawashima 2000-07-20 18:09 ` Maciej W. Rozycki 2000-07-19 14:54 ` J. Scott Kasten 2000-07-20 12:27 ` Simple Linux/MIPS 0.2b Ralf Baechle 2 siblings, 1 reply; 23+ messages in thread From: Hiroshi Kawashima @ 2000-07-19 14:19 UTC (permalink / raw) To: Maciej W. Rozycki Cc: Ralf Baechle, J. Scott Kasten, Keith M Wesolowski, linux-mips Hi. In message <Pine.GSO.3.96.1000719160110.21239D-100000@delta.ds2.pg.gda.pl>you w rite: > On Fri, 14 Jul 2000, Ralf Baechle wrote: > > > We have various known problems with the various binutils version around. > > We're working on getting a current snapshot of binutils working > > properly but right now we still have various problems, therefore > > egcs 1.0.3a + binutils 2.8.1 is still the recommended version. > > I have binutils 2.10 and gcc 2.95.2 which appear to be stable, i.e. I > haven't observed any problems recently. I may publish patches if anyone > is interested. I may see if I can arrange to publish RPM packages as > well. Great! Yes, I'm very interested in it. (Although I'm mipsel (linux-vr) guy, but it will be very helpful.) ---- Kawashima ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:19 ` Hiroshi Kawashima @ 2000-07-20 18:09 ` Maciej W. Rozycki 0 siblings, 0 replies; 23+ messages in thread From: Maciej W. Rozycki @ 2000-07-20 18:09 UTC (permalink / raw) To: Hiroshi Kawashima Cc: Ralf Baechle, J. Scott Kasten, Keith M Wesolowski, linux-mips On Wed, 19 Jul 2000, Hiroshi Kawashima wrote: > Yes, I'm very interested in it. OK, they are at 'http://www.ds2.pg.gda.pl/~macro/'. > (Although I'm mipsel (linux-vr) guy, but it will be very helpful.) Well, I'm working on a DECstation which is mipsel-linux, so results should be comparable. I suppose these fixes work for mips-linux equally well. I'll see if I can arrange space for RPMs, yet. -- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available + ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:04 ` Maciej W. Rozycki 2000-07-19 14:19 ` Hiroshi Kawashima @ 2000-07-19 14:54 ` J. Scott Kasten 2000-07-19 16:28 ` Andy Isaacson ` (2 more replies) 2000-07-20 12:27 ` Simple Linux/MIPS 0.2b Ralf Baechle 2 siblings, 3 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-19 14:54 UTC (permalink / raw) To: Maciej W. Rozycki; +Cc: linux-mips, linuxce-devel On Wed, 19 Jul 2000, Maciej W. Rozycki wrote: > I have binutils 2.10 and gcc 2.95.2 which appear to be stable, i.e. I > haven't observed any problems recently. I may publish patches if anyone > is interested. I may see if I can arrange to publish RPM packages as > well. Please do. There's many of us that would be interested. I for one am having zero success. I'm cross posting this from the SGI list to the linuxce list because even though one is big endian and the other is little endian, we all still use the same tools, and I simply need help or suggestions here regardless of the source. The short explination of what I'm doing is that I work for a company that wishes to use NEC vr or QED mips processors in some products using linux as the "embedded OS". I put quoates on that because our product environment is much fuller than the traditional embedded environment, to the point that we actually need a full blown *nix running under the hood. We plan to run the embedded processors big endian. We are using some SGI Indy boxes as development/eval stations at the moment as we evaluate porting our code base. The nice thing about the Indy is that the linux port and user land stuff is still all compiled 32 bit R3000 style, meaning that its binary compatible with our embedded systems, but a hell of a lot nicer to work on. We already know that user space binaries that run on the Indy work identically within our embeded environment. That much is cool. However, the problem is that I am having only limited success compiling user space stuff myself on the Indy, be it our own code, or any of the common distribution libs/apps. I cannot compile shared objects of good size and expect apps built from them to work. Xlibs, lesstif, etc and apps built against them just bus error. However if I go to one of the sites and tear open a prebuilt rpm for xlibs, or what ever, I can build against that and get an app to work just fine, but I cannot build the libraries themselves. I've tried every version of gcc/egcs/binutils that's supposed to work, and in every combination. However, I'm just not able to duplicate the kind of success that others appear to be having. Smaller things usually work. I've gotten sources from CVS, applied patches, even tried cross compiling, all to no avail. Unfortunately, it's critical that I be able to build things like that. Anyone that's having good success, please give me some info on what your build system/process looks like. I'm completely clueless why I'm not having the level of success that others appear to be having. Started with Simple Linux 0.1, glibc-2.0.6, kernel 2.2.14, egcs-1.1.2, binutils 2.9.5, on Indy 128M ram, R4400SC 200MHz, XL-24. Have tried the "manhattan" or "hard hat 5.1" SGI distro, but have never been able to get that to install. The 2.1.100 kernel included panics on the Indy due to page faults in an interupt handler (says so on the screen dump). Tried swapping that out for the 2.2.14 kernel from Simle Linux which runs very stable on that box, but there appears to be a kernel/glibc mismatch at that point because things don't work right after the boot. Would like to try the debian distro, but cannot find a big endian root image to install with, the site with the eb pacakages seems to be down, and cannot find the userland utility to unpack the .deb files, so this one is out completely at the moment. Have tried native compiles with binutils 2.8.1, egcs-1.0.2, gcc-2.7.2. The gcc-2.7.2/bin-2.8.1 combo was the closest that I've come. Was able to comple lesstif against Xlibs stolen from Hard Hat 5.1 distro and the programs in the test directory all seemed to work. However, could not get libs built from XFree86 4.0.1 or CVS to work. Builds, but everything bus errors. Unfortunately, our c++ apps will not compile with g++-2.7.2 even if that did work. To clarify, I have no interest in the XServer, just the xlibs so I can throw output out over the network. Have also tried cross compiling using binutils 2.8.1 and egcs-1.0.3a on an x86 box, but still same results. Little things work, big things just bus error. Thanks for any sugestions. I'm certainly stumped at this point. ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:54 ` J. Scott Kasten @ 2000-07-19 16:28 ` Andy Isaacson 2000-07-19 17:13 ` Keith M Wesolowski 2000-07-26 15:53 ` big endian Debian root image Chris Ruvolo 2 siblings, 0 replies; 23+ messages in thread From: Andy Isaacson @ 2000-07-19 16:28 UTC (permalink / raw) To: J. Scott Kasten; +Cc: Maciej W. Rozycki, linux-mips, linuxce-devel On Wed, Jul 19, 2000 at 11:54:59AM -0300, J. Scott Kasten wrote: > try the debian distro, but cannot find a big endian root image to install > with, the site with the eb pacakages seems to be down, and cannot find the > userland utility to unpack the .deb files, so this one is out completely > at the moment. A .deb file is just an ar(1) archive. Use "ar xv foo.deb" to unpack. -andy -- Andy Isaacson <adi@mr-happy.com> http://web.mr-happy.com/~adi/ ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:54 ` J. Scott Kasten 2000-07-19 16:28 ` Andy Isaacson @ 2000-07-19 17:13 ` Keith M Wesolowski 2000-07-19 17:42 ` J. Scott Kasten 2000-07-19 22:56 ` J. Scott Kasten 2000-07-26 15:53 ` big endian Debian root image Chris Ruvolo 2 siblings, 2 replies; 23+ messages in thread From: Keith M Wesolowski @ 2000-07-19 17:13 UTC (permalink / raw) To: J. Scott Kasten; +Cc: Maciej W. Rozycki, linux-mips, linuxce-devel On Wed, Jul 19, 2000 at 11:54:59AM -0300, J. Scott Kasten wrote: > However, the problem is that I am having only limited success compiling > user space stuff myself on the Indy, be it our own code, or any of the > common distribution libs/apps. I cannot compile shared objects of good > size and expect apps built from them to work. Xlibs, lesstif, etc and > apps built against them just bus error. However if I go to one of the This is interesting. I have seen plenty of problems (with several toolchains) building pic objects and then using -static to link. But linking against shared objects seems to work fine. I have successfully built the X libs (4.0 + Guido's newport patch) with the toolchain included in Simple 0.2b, and have run applications linked against them. The entire system was built from the same toolchain. > swapping that out for the 2.2.14 kernel from Simle Linux which runs very > stable on that box, but there appears to be a kernel/glibc mismatch at > that point because things don't work right after the boot. Would like to Could you explain in greater detail? For example, the userland in Simple 0.2b will not work with kernel < 2.3.51, and will give an error to that effect if you try to boot it with 2.2.14. This is intentional; it seems for some reason that glibc 2.2 userland on a 2.2 kernel will have, actually, the kind of behaviour you are observing. > errors. Unfortunately, our c++ apps will not compile with g++-2.7.2 even > if that did work. And unfortunately we still don't have a working c++ compiler for 2.96. Hopefully we're close though. I almost wonder if your hardware is broken somehow, as the system you describe is identical in every way to the one I am using to develop Simple. -- Keith M Wesolowski wesolows@chem.unr.edu University of Nevada http://www.chem.unr.edu Chemistry Department Systems and Network Administrator ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 17:13 ` Keith M Wesolowski @ 2000-07-19 17:42 ` J. Scott Kasten 2000-07-19 22:56 ` J. Scott Kasten 1 sibling, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-19 17:42 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips, linuxce-devel On Wed, 19 Jul 2000, Keith M Wesolowski wrote: > On Wed, Jul 19, 2000 at 11:54:59AM -0300, J. Scott Kasten wrote: > > However, the problem is that I am having only limited success compiling > > user space stuff myself on the Indy, be it our own code, or any of the > > common distribution libs/apps. I cannot compile shared objects of good > > size and expect apps built from them to work. Xlibs, lesstif, etc and > > apps built against them just bus error. However if I go to one of the > > This is interesting. I have seen plenty of problems (with several > toolchains) building pic objects and then using -static to link. But > linking against shared objects seems to work fine. I have successfully > built the X libs (4.0 + Guido's newport patch) with the toolchain > included in Simple 0.2b, and have run applications linked against > them. The entire system was built from the same toolchain. I pulled down 0.2b yesterday an will be giving that a whirl today of course. However, yes, this problem is interesting. I've used GDB just enough on small apps compiled against xlibs/lesstif (things in the lesstif test directory) to gather limited info on the problem. Some calls into the libraries work, however some fail. I've not isolated the point of failure, but when it does, the program counter is way off in impossible memory, and gdb has lost the program call stack so it cannot tell me where I am, or how I got there. This behavior is not random, it appears that it's exactly reproduceable, so with sufficient effort could presumably be tracked down. However, my pressing need is simply to get a tool chain that works within reason. Although I'm interested in doing some general sluthing to do my part to improve the quality of publicly available tools, unfortunately that's a luxury that must wait a little while when you have corporate deadlines to meet. > > swapping that out for the 2.2.14 kernel from Simle Linux which runs very > > stable on that box, but there appears to be a kernel/glibc mismatch at > > that point because things don't work right after the boot. Would like to > > Could you explain in greater detail? For example, the userland in > Simple 0.2b will not work with kernel < 2.3.51, and will give an > error to that effect if you try to boot it with 2.2.14. This is > intentional; it seems for some reason that glibc 2.2 userland on a 2.2 > kernel will have, actually, the kind of behaviour you are observing. Simple 0.1 has run very stable for me (meaning no crashes, lockups, etc...) using the 2.2.14 kernel. The only problem I have is building my own shared objects. I've beat that system for days on end with large compiles, heavy NFS traffic, etc. and not had it mis-behave. I tried booting the 5.1 install image using the included kernel (2.1.100 I beleive), but it either hangs after initializing eth0, or panics sometime there after with a "page fault in irq handler" message. On a whim, I just plunked down 2.2.14 which I know runs smooth and tried booting the 5.1 install. I then get a message "Cannot open initial console." Yes, the dev nodes are in place and so forth. Some time ago, I had gotten past that, but don't remember how. Anyway, the result was that things like fdisk could not write the disk and on and on. This sounds like a glibc missmatch. I could try slipping that in too and see what happens. In order to move quickly, I had pulled a few libraries from the 5.1 and plunked them down under Simple to try some things, having checked dependencies and so forth of course. I fully admit that such a move could indeed be part of my problem. However it's the libraries that I've stolen in this fashion that work! Anyway, point is, I've tried moving to a more complete distro where things have been uniformly compiled with the same tool chain to see if the shared object problem goes away. However, I've not been successfull in that endeavor either. > > errors. Unfortunately, our c++ apps will not compile with g++-2.7.2 even > > if that did work. > > And unfortunately we still don't have a working c++ compiler for > 2.96. Hopefully we're close though. I'd be happy to use different c and c++ compilers, if it worked well. However, it is of course preferable to remain consistent. There would even be an interest in modifying our own C++ code so that templates would get through g++ 2.7.2 if we had too. However, it's got to work before we can make that effort. > I almost wonder if your hardware is broken somehow, as the system you > describe is identical in every way to the one I am using to develop > Simple. Beleive me, the thought has crossed my mind. We have 3 identical boxes that were distributed through our different project teams. I will be trying this on one of those minimum as well. Thanks again for your assistance. -JSK- ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 17:13 ` Keith M Wesolowski 2000-07-19 17:42 ` J. Scott Kasten @ 2000-07-19 22:56 ` J. Scott Kasten 2000-07-20 1:39 ` Keith M Wesolowski 2000-07-20 1:51 ` Keith M Wesolowski 1 sibling, 2 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-19 22:56 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips On Wed, 19 Jul 2000, Keith M Wesolowski wrote: > I have successfully > built the X libs (4.0 + Guido's newport patch) with the toolchain > included in Simple 0.2b, and have run applications linked against > them. The entire system was built from the same toolchain. There is one thing at this point that would really really help. I need to establish a reference point and follow someone elses procedures exactly to determine what the source of my problem is and why some of you are not seeing this to the extent that I am. In this way, we can verify whether or not there is a bonified problem, and if it is with the tools, maybe get it fixed. I think the x libs are a good place to start, because I can guarantee that I'm breaking those, yet you claim to have them working. I also know that I can grab someone else's prebuilt and compile against those and get working apps. Thus the x libs demonstrate all the "features" of the problem that I'm encountering. If you can answer some questions, I can try to parallel what you have done: #1 You say you built X using the tool chain for Simple 0.2b. Does that mean you compiled it natively on an Indy with what was effectively Simple 0.2b installed? Or was it cross compiled using the same releases of the tools? #2 You say X 4.0. Can you be more specific? Was it an official 4.0 tarball on the xfree ftp site? Pulled from CVS? Or was it really a 4.0.1? Basicly, I want the exact same starting sources that you used. #3 Where can I get Guido's X patch? #4 What if anything did you add to the host.def or other config files before doing "make World"? ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 22:56 ` J. Scott Kasten @ 2000-07-20 1:39 ` Keith M Wesolowski 2000-07-20 1:19 ` J. Scott Kasten 2000-07-20 1:51 ` Keith M Wesolowski 1 sibling, 1 reply; 23+ messages in thread From: Keith M Wesolowski @ 2000-07-20 1:39 UTC (permalink / raw) To: J. Scott Kasten; +Cc: linux-mips On Wed, Jul 19, 2000 at 07:56:25PM -0300, J. Scott Kasten wrote: > I think the x libs are a good place to start, because I can guarantee that > I'm breaking those, yet you claim to have them working. I also know that > I can grab someone else's prebuilt and compile against those and get > working apps. Thus the x libs demonstrate all the "features" of the > problem that I'm encountering. Sure. > #1 You say you built X using the tool chain for Simple 0.2b. Does that > mean you compiled it natively on an Indy with what was effectively Simple > 0.2b installed? Or was it cross compiled using the same releases of the > tools? It was built natively. You will need to obtain cpp from egcs 1.1.2 and place it into /lib in order to accomplish this. You can get this at ftp://oss.sgi.com/pub/linux/mips/mips-linux/simple/userland-0.2b/bin/cpp > #2 You say X 4.0. Can you be more specific? Was it an official 4.0 > tarball on the xfree ftp site? Pulled from CVS? Or was it really a > 4.0.1? Basicly, I want the exact same starting sources that you used. It is XFree86 4.0 official + the 000531 version of Guido's patch. > #3 Where can I get Guido's X patch? ftp://oss.sgi.com/pub/linux/mips/X/source/newport_000531.diff.gz > #4 What if anything did you add to the host.def or other config files > before doing "make World"? I changed nothing. -- Keith M Wesolowski wesolows@chem.unr.edu University of Nevada http://www.chem.unr.edu Chemistry Department Systems and Network Administrator ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-20 1:39 ` Keith M Wesolowski @ 2000-07-20 1:19 ` J. Scott Kasten 0 siblings, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-20 1:19 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips Great. I'll get in sync with that and see where it takes me. I did try recompiling our own C++ code on one of the other Indy boxes to check out the possibility of a hardware problem. That one was also running Simple 0.1. It exhibeted the same bus error. This alone is not conclusive one way or another. That code was developed on x86 and I don't know if that group ever tested it on Sparc or other risc for endian/alignment issues in the data structures. Thus any combination of the code, compiler, machine could be broken. Further analysis will tell. My company seems comitted to using mips, as opposed to moving to any other arch as I was told to "make it work regardless" this afternoon, even if it meant changeing their time line a little. They seem interested in throwing some people at this if it comes down to fixing the tools to get the job done. For arguments sake, if that were the case, what should we be working with and what people on the mailing lists should I start getting in touch with? Thanks again for your input Keith! ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 22:56 ` J. Scott Kasten 2000-07-20 1:39 ` Keith M Wesolowski @ 2000-07-20 1:51 ` Keith M Wesolowski 2000-07-20 1:43 ` J. Scott Kasten 1 sibling, 1 reply; 23+ messages in thread From: Keith M Wesolowski @ 2000-07-20 1:51 UTC (permalink / raw) To: J. Scott Kasten; +Cc: linux-mips On Wed, Jul 19, 2000 at 07:56:25PM -0300, J. Scott Kasten wrote: > There is one thing at this point that would really really help. I need to > establish a reference point and follow someone elses procedures exactly to > determine what the source of my problem is and why some of you are not > seeing this to the extent that I am. In this way, we can verify whether > or not there is a bonified problem, and if it is with the tools, maybe get > it fixed. Before you get carried away with this, I should point out that I've been looking at this problem in more detail. Many of my X programs don't work either; they get SEGV, never bus errors. I suspect this is a glibc bug. I can run a few programs, such as xhost and the server itself. I did have a full X setup, including a working server, under 0.1. That was built exactly the same way I've done it under 0.2b. You may wish to try both and see which works better. If you like, I will return to 0.1 and duplicate my previous efforts. I can then describe the exact procedure to you. -- Keith M Wesolowski wesolows@chem.unr.edu University of Nevada http://www.chem.unr.edu Chemistry Department Systems and Network Administrator ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-20 1:51 ` Keith M Wesolowski @ 2000-07-20 1:43 ` J. Scott Kasten 0 siblings, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-20 1:43 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips On Wed, 19 Jul 2000, Keith M Wesolowski wrote: > On Wed, Jul 19, 2000 at 07:56:25PM -0300, J. Scott Kasten wrote: > > > There is one thing at this point that would really really help. I need to > > establish a reference point and follow someone elses procedures exactly to > > determine what the source of my problem is and why some of you are not > > seeing this to the extent that I am. In this way, we can verify whether > > or not there is a bonified problem, and if it is with the tools, maybe get > > it fixed. > > Before you get carried away with this, I should point out that I've > been looking at this problem in more detail. Many of my X programs > don't work either; they get SEGV, never bus errors. I suspect this is > a glibc bug. I can run a few programs, such as xhost and the server > itself. I did have a full X setup, including a working server, under > 0.1. That was built exactly the same way I've done it under 0.2b. You > may wish to try both and see which works better. Alright, that's starting to sound more like what I'm seeing. Some things work, some don't. Some SEGV, some bus error. The bugs float from one place to another depending on what tools I used to build. My litimus test has been to launch an xterm over the network, xeyes, and xwininfo. If those work, then I proceed to building lesstif against that set. Unfortunately, only libraries that others have built have gotten me to the lesstif stage. In fact, I've even had difficulty trying to build the compiler itself natively. With egcs-1.03a (patched) it would not natively build using the egcs-1.1.2, or gcc-2.7.2. Basicly at the point where the build starts launching those "genXXXXX" functions, they start segv or bus erroring. The ones that break depend on which compiler was used. Thus I was forced to try that one as a cross compiler, although I just found an rpm for it so I can finally give it a whirl natively. > If you like, I will return to 0.1 and duplicate my previous efforts. I > can then describe the exact procedure to you. Thanks, but not necessary just yet. I'll try using the X sources and patch you mentioned on both Simple 0.1 and Simple 0.2b and see what happens. We can then compare notes to see what workes, what doesn't. At that point I'm confident that we can rule out a lot of the variables and get down to demonstrating specific problems with the tools themselves. Also at that point, I can start stuffing the broken binaries into one of our eval boards and have our hardware people run it against a logic analyser so we can get to the bottom of some of these things quickly if they'll let me have the resources. We're having a staff meeting in the morning about exactly that. ;-) ^ permalink raw reply [flat|nested] 23+ messages in thread
* big endian Debian root image 2000-07-19 14:54 ` J. Scott Kasten 2000-07-19 16:28 ` Andy Isaacson 2000-07-19 17:13 ` Keith M Wesolowski @ 2000-07-26 15:53 ` Chris Ruvolo 2000-07-26 17:10 ` J. Scott Kasten 2 siblings, 1 reply; 23+ messages in thread From: Chris Ruvolo @ 2000-07-26 15:53 UTC (permalink / raw) To: J. Scott Kasten; +Cc: linux-mips On Wed, 19 Jul 2000, J. Scott Kasten wrote: > Would like to >try the debian distro, but cannot find a big endian root image to install >with, I've put one together. Its rather limited and a bit outdated now (doesn't have perl 5.005 nor debconf), but you might want to take a look at it. It has a working dpkg & apt. I plan on updating this within the next week or two. http://ftp.rfc822.org/pub/local/debian-mips/experimental/debian-mips-base-test_0.3.tar -Chris ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: big endian Debian root image 2000-07-26 15:53 ` big endian Debian root image Chris Ruvolo @ 2000-07-26 17:10 ` J. Scott Kasten 0 siblings, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-26 17:10 UTC (permalink / raw) To: Chris Ruvolo; +Cc: linux-mips Thanks. I'll give it a shot in the next few days. On Wed, 26 Jul 2000, Chris Ruvolo wrote: > On Wed, 19 Jul 2000, J. Scott Kasten wrote: > > > Would like to > >try the debian distro, but cannot find a big endian root image to install > >with, > > I've put one together. Its rather limited and a bit outdated now (doesn't > have perl 5.005 nor debconf), but you might want to take a look at it. > It has a working dpkg & apt. I plan on updating this within the next week > or two. > > http://ftp.rfc822.org/pub/local/debian-mips/experimental/debian-mips-base-test_0.3.tar > > -Chris > > > > ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-19 14:04 ` Maciej W. Rozycki 2000-07-19 14:19 ` Hiroshi Kawashima 2000-07-19 14:54 ` J. Scott Kasten @ 2000-07-20 12:27 ` Ralf Baechle 2000-07-20 18:57 ` Maciej W. Rozycki 2000-07-21 10:21 ` Maciej W. Rozycki 2 siblings, 2 replies; 23+ messages in thread From: Ralf Baechle @ 2000-07-20 12:27 UTC (permalink / raw) To: Maciej W. Rozycki Cc: J. Scott Kasten, Keith M Wesolowski, linux-mips, linux-mips, linux-mips On Wed, Jul 19, 2000 at 04:04:58PM +0200, Maciej W. Rozycki wrote: > > We have various known problems with the various binutils version around. > > We're working on getting a current snapshot of binutils working > > properly but right now we still have various problems, therefore > > egcs 1.0.3a + binutils 2.8.1 is still the recommended version. > > I have binutils 2.10 and gcc 2.95.2 which appear to be stable, i.e. I > haven't observed any problems recently. I may publish patches if anyone > is interested. I may see if I can arrange to publish RPM packages as > well. Anyway, have you tried to rebuild an entire Linux distribution with your toolset, including kernel, static (Like rpm or ldconfig) and dynamic PIC code? If that was successful I'd really like to recommend your toolset as the new standard - especially binutils 2.8.1 are just too rotten. Example: [ralf@indy /tmp]$ echo 'main(){}' > c.c;gcc -o c c.c -lm -lieee collect2: ld terminated with signal 11 [Segmentation fault], core dumped This nukes binutils 2.8.1 reliably ... The problem is extracted from the INN build but also affects a few more packages. Ralf ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-20 12:27 ` Simple Linux/MIPS 0.2b Ralf Baechle @ 2000-07-20 18:57 ` Maciej W. Rozycki 2000-07-21 10:21 ` Maciej W. Rozycki 1 sibling, 0 replies; 23+ messages in thread From: Maciej W. Rozycki @ 2000-07-20 18:57 UTC (permalink / raw) To: Ralf Baechle Cc: J. Scott Kasten, Keith M Wesolowski, linux-mips, linux-mips, linux-mips On Thu, 20 Jul 2000, Ralf Baechle wrote: > Anyway, have you tried to rebuild an entire Linux distribution with your No I didn't and I don't expect anytime soon. Basically due to the lack of disk space and time to tweak broken RPMs. I've effectively build 68 of different binary RPMs so far (not counting different releases of the same packages) and these include tools important to run a minimal system as well as have a reasonable development environment. These include SysVinit 2.78, bash 2.04, binutils 2.10, db 3.0.55, gcc 2.95.2, gdb 5.0, glibc 2.1.91, pam 0.72, rpm 3.0.4, and others. > toolset, including kernel, static (Like rpm or ldconfig) and dynamic PIC > code? If that was successful I'd really like to recommend your toolset As far as I can see both statically (bash.static, rpm, ldconfig) and dynamically linked code (almost everything else) works. Dlopening seems to work as NSS and PAM modules appear to work but it needs further investigation. Every recent failure I observed was a result of a bug in glibc or the kernel. > as the new standard - especially binutils 2.8.1 are just too rotten. Agreed. > Example: > > [ralf@indy /tmp]$ echo 'main(){}' > c.c;gcc -o c c.c -lm -lieee > collect2: ld terminated with signal 11 [Segmentation fault], core dumped This works for me with a cross-compiler from an i386 host. I'll try it natively at home tonight. I don't expect a failure, though. Gcc is rock-solid for me for quite some time, and the last failure of binutils I've seen was the problem of relocations in .data section in an executable I've discussed with Ulf at the end of May. Gcc 2.96 is just to unstable to be used unless developing it and there is no point in using binutils 2.10.90 for most people. > This nukes binutils 2.8.1 reliably ... The problem is extracted from the > INN build but also affects a few more packages. I will build packages progressively but bugfixes and improvements are higher on my TODO list than a working distribution -- I can work with init starting bash on the console. ;-) I hope to have working libpthreads tonight (fixes are in yesterday's changes and I have a patch to support R3K) -- this is the last remaining obstacle that prevents me from being able to build .mipsel.rpm packages natively (not all software cross-builds well; glibc included). I'll be able to test various software more thoroughly then (though building glibc over NFS would certainly not be running at a stunning speed, sigh...). I will definitely work for getting some FTP space for RPMs. Maciej -- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available + ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-20 12:27 ` Simple Linux/MIPS 0.2b Ralf Baechle 2000-07-20 18:57 ` Maciej W. Rozycki @ 2000-07-21 10:21 ` Maciej W. Rozycki 1 sibling, 0 replies; 23+ messages in thread From: Maciej W. Rozycki @ 2000-07-21 10:21 UTC (permalink / raw) To: Ralf Baechle Cc: J. Scott Kasten, Keith M Wesolowski, linux-mips, linux-mips, linux-mips On Thu, 20 Jul 2000, Ralf Baechle wrote: > [ralf@indy /tmp]$ echo 'main(){}' > c.c;gcc -o c c.c -lm -lieee > collect2: ld terminated with signal 11 [Segmentation fault], core dumped The above example compiles fine for me natively and the resulting binary works (i.e. it doesn't crash; what else would anyone expect from it? ;-) ). Also I built a CVS snapshot of glibc 2.1.91 taken Jul 20th, 2000 at 2:25 UTC and it solves the libpthreads showstopper as I expected. Maciej -- + Maciej W. Rozycki, Technical University of Gdansk, Poland + +--------------------------------------------------------------+ + e-mail: macro@ds2.pg.gda.pl, PGP key available + ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 13:58 ` J. Scott Kasten 2000-07-13 22:51 ` Ralf Baechle @ 2000-07-13 23:30 ` Keith M Wesolowski 2000-07-13 22:59 ` J. Scott Kasten 1 sibling, 1 reply; 23+ messages in thread From: Keith M Wesolowski @ 2000-07-13 23:30 UTC (permalink / raw) To: J. Scott Kasten; +Cc: Keith M Wesolowski, linux-mips On Thu, Jul 13, 2000 at 10:58:48AM -0300, J. Scott Kasten wrote: > already verified that the binaries are portable.) However, I am having > difficulty on the Indy itself. NONE of the shared objects that I build on > that box can be used by programs on that box. Yes, I either placed them It's not clear why that is. For example, ncurses is a shared library, and bash is linked against it. That works fine. > I've tried building the XFree 4.0.1 libraries, Lesstiff libraries, and > libraries from our own code base. All breaks the same way. I have successfully built XFree on 0.1. It even worked. > #2 Are there specific compiler flags/phases that should/shouldn't be used > with MIPS arch so's that are different than what I'd normally do under > Linux? No. > #3 Should I be trying to cross compile the so's them selves instead of > native builds? If anything that would be worse. It might be faster, though. :) > #4 Are you using different flavors of gcc/binutils to do different jobs > because of known breakages? I'm not because I feel that's a cop-out. If it doesn't work, we need to fix it, not just keep using old stuff forever as a band-aid. I don't mean offense to anyone; obviously many people need stability rather than currency so their take on this will differ from mine; I just don't believe that using the old toolchain is the right approach in the long run. It's important to get the -current stuff working and the only way that will happen is if people find and fix bugs. There are already a few waiting for some intrepid volunteer to fix; see the 0.2b release notes and various mailing lists. There are known breakages with binutils, for all versions I am aware of, and for most if not all gcc versions. However, for shared libraries and dynamic linking, the versions included in 0.2b seem to work fine. Ideally the new toolchain will have _better_ quality than the old one very soon; when that does, everyone can switch and the ancient stuff can finally die. In the meantime, consider everything I announce to be of experimental quality, in case I haven't been adequately clear thus far. In the meantime, Ralf and a number of others are using egcs 1.0.3a and binutils 2.8.1, both with patches. These may be more reliable for you. > Any tips appreciated here. In the mean, I'm going to start playing with > flags. Please don't use 0.1 any more. If you want a glibc 2.0 system, use Hard Hat or some other "officially sanctioned" distribution. Use 0.2b if you want to play with the current stuff. Sorry for any confusion. -- Keith M Wesolowski wesolows@chem.unr.edu University of Nevada http://www.chem.unr.edu Chemistry Department Systems and Network Administrator ^ permalink raw reply [flat|nested] 23+ messages in thread
* Re: Simple Linux/MIPS 0.2b 2000-07-13 23:30 ` Keith M Wesolowski @ 2000-07-13 22:59 ` J. Scott Kasten 0 siblings, 0 replies; 23+ messages in thread From: J. Scott Kasten @ 2000-07-13 22:59 UTC (permalink / raw) To: Keith M Wesolowski; +Cc: linux-mips On Thu, 13 Jul 2000, Keith M Wesolowski wrote: > > However, I am having > > difficulty on the Indy itself. NONE of the shared objects that I build on > > that box can be used by programs on that box. Yes, I either placed them > > It's not clear why that is. For example, ncurses is a shared library, > and bash is linked against it. That works fine. I know, all the other SO's work too, it only seems to affect the ones that I build, however, I have discovered that some of them do work, but many don't. > > I've tried building the XFree 4.0.1 libraries, Lesstiff libraries, and > > libraries from our own code base. All breaks the same way. > > I have successfully built XFree on 0.1. It even worked. Well, that's the status that I'm aiming for! > > #3 Should I be trying to cross compile the so's them selves instead of > > native builds? > > If anything that would be worse. It might be faster, though. :) Agreed, however I'm just looking for a solution. > > #4 Are you using different flavors of gcc/binutils to do different jobs > > because of known breakages? > > I'm not because I feel that's a cop-out. If it doesn't work, we need > to fix it, not just keep using old stuff forever as a band-aid. I > don't mean offense to anyone; obviously many people need stability > rather than currency so their take on this will differ from mine; I > just don't believe that using the old toolchain is the right approach > in the long run. It's important to get the -current stuff working and > the only way that will happen is if people find and fix bugs. There > are already a few waiting for some intrepid volunteer to fix; see the > 0.2b release notes and various mailing lists. There are known > breakages with binutils, for all versions I am aware of, and for most > if not all gcc versions. However, for shared libraries and dynamic > linking, the versions included in 0.2b seem to work fine. Ideally the > new toolchain will have _better_ quality than the old one very soon; > when that does, everyone can switch and the ancient stuff can finally > die. In the meantime, consider everything I announce to be of > experimental quality, in case I haven't been adequately clear thus > far. Again, I agree that is the aproach to take in the long run. However, my need is very urgent so I need to find something that works however I have to do it as I'm attempting to meet a corporate deadline for some development. > In the meantime, Ralf and a number of others are using egcs 1.0.3a and > binutils 2.8.1, both with patches. These may be more reliable for you. I will evaluate these as I evaluate the tools from the SLM 0.2b release as well. The catch I have with gcc-2.96 is that I must have the C++ compiler as well. But again, I'll figure out what works for me and what doesn't. Mainly, I needed to clarify that you guys were not doing anything special with compiler flags, etc... > Please don't use 0.1 any more. If you want a glibc 2.0 system, use > Hard Hat or some other "officially sanctioned" distribution. Use 0.2b > if you want to play with the current stuff. Well, other than the SO problem, I've had really good luck with that, and it's really jumpstarted my work quite far. I've stolen bits and pieces of what wasn't in there from Hard Hat as needed. ;-) I'll be setting up an 0.2b partition soon to evaluate the 2.4.0 kernel and updated glibc as well at some point. ^ permalink raw reply [flat|nested] 23+ messages in thread
end of thread, other threads:[~2000-07-26 17:13 UTC | newest] Thread overview: 23+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2000-07-13 7:16 Simple Linux/MIPS 0.2b Keith M Wesolowski 2000-07-13 13:58 ` J. Scott Kasten 2000-07-13 22:51 ` Ralf Baechle 2000-07-13 22:39 ` J. Scott Kasten 2000-07-19 14:04 ` Maciej W. Rozycki 2000-07-19 14:19 ` Hiroshi Kawashima 2000-07-20 18:09 ` Maciej W. Rozycki 2000-07-19 14:54 ` J. Scott Kasten 2000-07-19 16:28 ` Andy Isaacson 2000-07-19 17:13 ` Keith M Wesolowski 2000-07-19 17:42 ` J. Scott Kasten 2000-07-19 22:56 ` J. Scott Kasten 2000-07-20 1:39 ` Keith M Wesolowski 2000-07-20 1:19 ` J. Scott Kasten 2000-07-20 1:51 ` Keith M Wesolowski 2000-07-20 1:43 ` J. Scott Kasten 2000-07-26 15:53 ` big endian Debian root image Chris Ruvolo 2000-07-26 17:10 ` J. Scott Kasten 2000-07-20 12:27 ` Simple Linux/MIPS 0.2b Ralf Baechle 2000-07-20 18:57 ` Maciej W. Rozycki 2000-07-21 10:21 ` Maciej W. Rozycki 2000-07-13 23:30 ` Keith M Wesolowski 2000-07-13 22:59 ` J. Scott Kasten
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).