September 2004 - openSUSE Programming - openSUSE Mailing Lists

Device Driver problems For Nat Inst PCI-DIO-32HS Card
by Charles Hicks 27 Sep '04

27 Sep '04

Using SuSE 9.1 on a 32 bit machine. The only available Linux driver for the National Instruments PCI-DIO-32HS I/O Card appears to be the ni_pcidio driver from comedi.org. However, the DMA mode is not fully functional. It only works in a "timed input mode", and we need it to work in burst mode. Adding code to this driver to gain the burst mode functionality seems quite formidable to someone with very little device driver experience, given the myriad of data structures in the code. A second option we pursued was some Linux code from the National Instruments web site. The code to perform DMA transfers is there except for one small issue. There are function stubs bufAlloc and bufFree to perform the DMA memory allocation and freeing. I found some code in Chapter 7 of "The Linux Kernel Module Programming Guide" from the Linux Documentation Project that performed the following functionality: One is a LKM that copies a buffer from user space to a static buffer in kernel space. The other code runs in user space and uses a /dev/xxxx file and an ioctl function call to direct the kernel to copy the user space buffer to the kernel space buffer. There is also functionality in the latter to direct the kernel to copy a kernel space buffer into a user space buffer. I modified the LKM code to dynamically allocate and free a kernel buffer using pci_allocate_consistent and pci_free_consistent, and it works as described above. The problem is that the bufAlloc and bufFree stubs are called directly from user space in the NI code. I have tried modifying the NI code so that it uses the LKM with no success. My questions: Is the only way to allocate DMA memory in Linux with a LKM that runs in kernel space? Is the only way to access the DMA memory via a /dev/xxxx device? Any suggestions or pointers or even constructive comments would be greatly appreciated! Sorry for the long note!

1 0

Problems with POSIX Asynchronous IO
by Matthew Gregan 27 Sep '04

27 Sep '04

Hi all, I'm having some problems with the POSIX Asynchronous IO support on a SLES 9 box I've been working with. The machine has all of the available SLES 9 updates applied. After submitting a single IO using the POSIX AIO interface provided by glibc, I notice that the worker thread created by glibc enters a tight loop and consumes 100% of one CPU. Once this occurs, AIO continues to work fine (other than this problem), but the worker thread will remain in this state until the program exits. The kernel does not log any errors during the occurrence of this problem, and the machine has otherwise been running fine. I've written a small test case to reproduce the problem I'm seeing (attached as aiotest.c). Build the test case using 'cc -Wall -o aiotest aiotest.c -lrt' and run it--it'll create an 8MB file in the current directory, submit a single 128kB IO, and then sleep for 10 seconds. The test case is small and does little error checking; for the sake of this test case, I prefer exiting to handling any errors nicely. Since writing the test case, I've reproduced the problem on a number of other SLES 9 boxes, both x86 and z/Series. The same test case, when run on a SLES 8 box or RHEL 3AS box runs without issue, and the CPU is close to idle, as expected. I suspect the problem is within either the new kernel AIO library provided as /usr/lib/librtkaio.so.1, with the kernel, or an interaction between the two. RHEL 3AS also provides librtkaio for kernel AIO, and it seems to be a new addition to SLES 9 as it was not present on my SLES 8 test boxes. Running aiotest with LD_ASSUME_KERNEL=2.4.0 to force use of the older pre-NPTL glibc results in the userspace AIO interface provided by librt.so.1 being used, and the test case runs without consuming excessive CPU. Any suggestions would be appreciated. 'vmstat 1' during an aiotest run shows: procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu---- r b swpd free buff cache si so bi bo in cs us sy id wa 0 0 83356 399012 10168 61196 0 0 0 0 1012 55 0 0 100 0 0 0 83356 399012 10168 61196 0 0 0 0 1036 93 1 0 99 0 [aiotest starts] 2 0 83356 400204 10192 59508 0 0 0 788 1026 54722 20 29 51 0 [aiotest should be idle at this point] 1 0 83356 398476 10192 61252 0 0 0 0 1052 111790 36 64 0 0 1 0 83356 396684 10200 63000 0 0 0 0 1010 111877 38 62 0 0 1 0 83356 394892 10200 64752 0 0 0 0 1011 112199 37 63 0 0 1 0 83356 393164 10200 66504 0 0 0 0 1009 112273 39 61 0 0 1 1 83356 391372 10204 68232 0 0 0 8740 1071 110521 35 65 0 0 1 0 83356 389708 10212 69972 0 0 0 40 1019 111472 38 62 0 0 1 0 83356 387916 10216 71724 0 0 0 0 1008 112259 40 60 0 0 1 0 83356 386188 10216 73480 0 0 0 0 1009 112332 39 61 0 0 1 0 83356 384396 10220 75232 0 0 0 0 1011 112208 38 62 0 0 0 1 83356 383516 10220 76288 0 0 0 8372 1075 67572 25 36 21 18 [10 second sleep ends, aiotest exits] 0 0 83356 383580 10228 76288 0 0 0 40 1013 57 0 0 95 5 0 0 83356 383580 10228 76288 0 0 0 0 1010 52 1 0 99 0 A few seconds after starting aiotest: $ ps -e -L -o pcpu,pid,tid,comm,wchan' | grep [a]iotest 0.0 23316 23316 aiotest2 schedule_timeout 71.7 23316 23317 aiotest2 - Attaching with GDB and looking at the thread backtraces: Thread 2 (Thread 1077447600 (LWP 23317)): #0 0x400335b6 in handle_kernel_aio () from /lib/tls/librt.so.1 #1 0x401529ed in start_thread () from /lib/tls/libpthread.so.0 #2 0x400f09ca in clone () from /lib/tls/libc.so.6 Thread 1 (Thread 1075172000 (LWP 23316)): #0 0xffffe410 in ?? () #1 0xbfffef08 in ?? () #2 0x40149bb8 in __elf_set___libc_thread_subfreeres_element___rpc_thread_destroy__ () from /lib/tls/libc.so.6 #3 0xbfffed64 in ?? () #4 0x400c0696 in __nanosleep_nocancel () from /lib/tls/libc.so.6 #5 0x400c0473 in sleep () from /lib/tls/libc.so.6 #6 0x080486d7 in main () #0 0xffffe410 in ?? () Further observation of the worker thread indicates that it spins inside handle_kernel_aio(), which is part of kaio_misc.c in the rtkaio module of glibc. I've quickly compared the source of the rtkaio module from SLES 9 and a RHEL 3AS box. The source of the modules in each glibc is identical, which suggests that the difference between the working RHEL 3AS box and the broken SLES 9 box lies elsewhere in the glibc code, or within the kernel. Running the test case using 'strace -f' shows thousands syscalls like the one below. Compared to a run on a RHEL 3AS box, the number of these io_getevents() calls is abnormally high. Example output from 'strace -f ./aiotest': [...] [pid 23317] io_getevents(0x4017f000, 0, 0xa, 0x4038894c, 0) = 0 [...the line above is repeated thousands of times...] Cheers, -mjg -- Matthew Gregan |/ /| kinetik(a)orcon.net.nz

1 0

Q4 ppl/w both C++ and Java experience
by Steven T. Hatton 26 Sep '04

26 Sep '04

Q: This question is for people with both C++ and Java experience. Do you believe there are features of Java that C++ could also have, and would benefit from? My opinion: Typically when Java is praised in the presence of C++ programmers the person extolling any virtue of Java is ridiculed and insinuations about his or her intelligence and abilities are leveled against the person. I believe I am now in a position to speak with some authority regarding the two languages. I may not be one of the leading experts in either language, but I can program effectively in both of them. I firmly believe Java has certain features that make it easier to use, and faster to program in. Among those is a superior mechanism for locating and integrating libraries. In Java the `#include' and `using' declaration (or directive - but I don't do that) are united in the `import' statement. The 'interface' provided by a 'header' file and the 'object' code provided by *.o, *.a, *.la, etc., are combined into a single *.class file. The file is easily located by leveraging the most ubiquitous data storage and retrieval mechanisms in computing; the file system. In the C++ Standard ISO/IEC 14882:2003 there is no formal concept of a header file. All use of header files is by convention, and is based on the ancient artifact inherited from C called the C preprocessor. I know of no fundamental reason why C++ cannot use a mechanism similar to that of Java (and C#, BTW) for locating and importing resources. That one facility would probably increase my productivity to twice what it currently is, simply by freeing me of the task of maintaining mapping identifiers to external resources which often have no direct naming similarity, much less a deterministic relationsip between identifier name and file name. -- Regards, Steven

4 13

runtime problems between PIII & P4 ....
by William A. Mahaffey III 25 Sep '04

25 Sep '04

.... I work for a small engineering services company. We are currently using a LLNL Finite element code called Dyna3D to do some analyses on behalf of a client. We have a 1000 MHz PIII at work, running SuSE 8.2, all stock right off the CDROM. We compile the code under the Portland Group FORTRAN 77 compiler, pgf77, since all (!!!!) of the other compilers we have tried either won't compile the code or produce an executable which generates wrong answers, seg faults, or both. Last night I brought home a statically linked version of the Dyna3D executable & some cases to run on my 2400 MHz P4, also running SuSE 8.2, right off the CDROM. I set up a directory to work in & kicked off 10 cases, which should have taken about 30-35 hours to run. This A.M., all were done. 8 of the 10 finished with random seg faults, 1 successfully completed, & 1 had a program error termination (crushed elements, operator error ;-)). I am puzzled by the random seg faults. The P4 has a kernel called '2.4.20-64GB-SMP', while the PIII at work (& mine at home) have one called '2.4.20-4GB'. All 3 installs of SuSE 8.2 (my 933 MHz PIII, the 1000 MHz PIII at work, & my 2400 MHz P4 at home) are completely default, plain vanilla installs right off the CDROM, no recompiled kernels, no shenanighans. The P4 is NOT a multi-processor machine, just 1 CPU. Are there any known problems with this M.O. (compile code on PIII, run it on P4) under SuSE 8.2 ? Is there some problem with using the '...SMP' kernel on a single CPU machine (everything else seems to work fabulously) ? If so, why did the installer install the SMP kernel on my single CPU box ? I can provide any other info anyone wants (except the code, export restricted ;-)). Please advise, & thanks in advance for your time. William A. Mahaffey III Huntsville, Al. USA

1 0

Compiler WON'T work at ALL!!
by Steven Pasternak 24 Sep '04

24 Sep '04

whenever I try to './configure' anything I get this: bash: ./configure: /bin/sh: bad interpreter: Permission denied and when I do 'bash ./configure', 'sh ./configure', or even 'zsh ./configure' i get this: checking build system type... i686-pc-linux-gnu checking host system type... i686-pc-linux-gnu checking target system type... i686-pc-linux-gnu checking for a BSD-compatible install... /usr/bin/install -c checking for -p flag to install... yes checking whether build environment is sane... yes checking for gawk... gawk checking whether make sets $(MAKE)... yes checking for style of include used by make... GNU checking for gcc... gcc checking for C compiler default output... a.out checking whether the C compiler works... configure: error: cannot run C compiled programs. If you meant to cross compile, use `--host'. See `config.log' for more details. steven@linux:~/Documents/other/RPMs/SuSE/Look/kerastik-0.1> I use SuSE 9.1 kernel 2.6.4-52-default with gcc-3.3.3 (the one that came with SuSE.) I'm pretty sure that it is the compiler because i did 'export CC=~/bin/gcc' which was an empty text file, and got the same output except the line: checking for gcc... gcc was replaced with: checking for gcc... /home/steven/bin/cc and i got the same checking whether the C compiler works... configure: error: cannot run C compiled programs. If you meant to cross compile, use `--host'. See `config.log' for more details. stuff. what is wrong?

2 5

Re: [suse-programming-e] C++ curtiosity: default argument in a pointer to function.
by John Lamb 23 Sep '04

23 Sep '04

> On Monday 13 September 2004 17:05, John Lamb wrote: >> This compiles and works under gcc 3.3.3, but I suspect it's not >> legal int main() { double (*function)( double, double = 0 ) = >> ::function; > > My guess is you are wrong. Your code is right. A pointer is a scalar > value, and I don't believe the language cares what it points to when > it comes to initialization. I found the answer eventually, in footnote 84 of the C++ draft standard: pointers to functions and typedefs are not allowed to have default parameters. Since overloaded operators can have default parameters, it's possible to get nearly the same effect legally with (say) struct Function { double operator()( double, double = 0 ) const; static Function& instance(){ static Function* function; return *function; } private: Function(){}; }; ... Function& function = Function::instance(); But, should a language be this cryptic? -- JDL

1 0

bug in Eclipse 3.0.x GTK: text/font on button unreadable -- Fix (Suse 9.1)
by zabojh＠arcor.de 22 Sep '04

22 Sep '04

Hi, the default Eclipse 3.0.x GTK installation on Suse 9.1 might show (depending on the GTK version + theme) unreadable text on buttons which makes it difficult if not impossible to use Eclipse-GTK. If you apply the following fix, the buttons will be readable: (see also bugreport https://bugs.eclipse.org/bugs/show_bug.cgi?id=68568 ) Just add the following lines of text to a file listed in ENV variable GTK2_RC_FILES, e.g. to file $HOME/.kde/share/config/gtkrc : ############################################ style "eclipse-button" { GtkButton::default_border = {0,0,0,0} GtkButton::default_outside_border = {0,0,0,0} } class "*Button*" style : highest "eclipse-button" widget_class "*Button*" style : highest "eclipse-button" ############################################# Start Eclipse and the buttons will be readable :-) Jan

1 0

Java & Javax.swing
by Karl Wheeler 22 Sep '04

22 Sep '04

Noob question i'm afraid.. I'm attempting to learn Java and have installed JAVA from the Professional 9.1 disks. The tutorial prog I have been trying to use requires javax.swing, but after no errors at compile time i get a whole heap of errors when I run the program. I have no problem runnning Java code that just dumps text to screen. All of these errors are regarding javax.swing Where am I going wrong. All help, comments ( and flaming :) gratefully accepted Karl Wheeler

2 1

QPainter does not show up the image file
by Prabu Subroto 22 Sep '04

22 Sep '04

Dear my friends... I am using SuSE 9.1 and Qt 3.2 . I try this simple code: #include <qpixmap.h> #include <qpainter.h> void pixForm::init(QPainter * p) { p->drawPixmap( 0,0, QPixmap("images/iamlagiajan.jpg"),0,0,100,100); } No problem in compiling. But if I run the compiled code, no picture showed up in the dialog form. Where is my mistake? Please tell me. Thank you very much in advance. __________________________________ Do you Yahoo!? New and Improved Yahoo! Mail - Send 10MB messages! http://promotions.yahoo.com/new_mail

3 5

RE: [suse-programming-e] kdevelop
by Gustavo Dutra 21 Sep '04

21 Sep '04

> > > hi Gustavo, > > are you sure your project directory is writable for your > current username? > > -Mathieu > > Gustavo Dutra wrote: > > >Hi, > > > >I have been trying with no success to create a java kde project in > >kdevelop (kde 3.3 - suse 9.0) It aborts with "can't create file > >subdirs" message. > > > > Thanks for your answer. Yes, it is writeable, it's my home directory. I have no problem with other project types. In fact, no problem in debian sarge(tested yesterday).

1 0