Serial port programming problems
I run OpenBSD 4.6 (i386) on a PCEngines ALIX2c3, as a low power file/web/DHCP server. I would like to have this machine regularly retrieve data from an instrument which communicates over RS-232. I'm using a Prolific USB-RS232 converter (full dmesg for the ALIX below). I have no protocol documentation but have been able to reverse engineer a subset of the instrument's protocol. I have written a code to extract data from the instrument which compiles cleanly on OpenBSD, Linux (gcc) and Solaris x86 (Sun Studio). On Linux and Solaris, the resulting binary always reads the data that I want. On OpenBSD, the code appears to hang on every second execution, its entry in the output of top looking similar to: 0 204K 496K idle ttyin 0:00 0.00% reader I've come up with a minimal example to demonstrate where the problem begins, and have pasted it below. The code sends an initialisation string to the instrument and should get an ACK (ASCII decimal 6) back. On Linux it works; on OpenBSD, it works the first time after insertion of the USB-RS232 converter but not subsequently. I don't believe this is likely to be a hardware problem with the USB-RS232 converter, since I successfully used the same one in Linux and Solaris. I also think it unlikely to be the USB host controller, since I have identical behaviour on another OpenBSD box, with a different USB controller, see lines from its dmesg: uhci0 at pci0 dev 7 function 2 "Intel 82371AB USB" rev 0x01: irq 11 usb0 at uhci0: USB revision 1.0 uhub0 at usb0 "Intel UHCI root hub" rev 1.00/1.00 addr 1 uplcom0 at uhub0 port 2 "Prolific Technology Inc. USB-Serial Controller D" rev 1.10/4.00 addr 2 ucom0 at uplcom0 Running on OpenBSD, either on the ALIX or on my other box, the behaviour I see is: # cc -Wall -o ex_obsd ex_obsd.c # ./ex_obsd; sleep 5; ./ex_obsd; sleep 5; ./ex_obsd; sleep 5; ./ex_obsd 1 bytes available, read: 6 1 bytes available, read: 10 Expecting , got 10 1 bytes available, read: 10 Expecting , got 10 1 bytes available, read: 10 Expecting , got 10 Running identical code (save for a change of device name) on Linux (on another different machine), I get: $ ./ex_lin; sleep 5; ./ex_lin; sleep 5; ./ex_lin; sleep 5; ./ex_lin 1 bytes available, read: 6 1 bytes available, read: 6 1 bytes available, read: 6 1 bytes available, read: 6 On OpenBSD, I can get the desired behaviour if I manually unplug and re-plug the USB to serial converter during each sleep. I'd be very grateful if anyone has any advice on what I should change to get the desired behaviour from this code on OpenBSD. Many thanks in advance. Neil My code: #include #include #include #include #include #include #include int open_port(void);/* Opens port, returns file descriptor */ void set_port(int fd); /* Sets port options on file descriptor fd */ int main(void) { int n,fd; char resp; char initString[] = {0x0D, 0x01, 0x39, 0x31, 0x38}; char ack = 0x06; char eot = 0x03; fd = open_port(); set_port(fd); if(DEBUG != 0) printf("Port has been opened and set up\n"); // send 918 n = write(fd,initString,5); if (n != 5) {printf("Failed to write init string.\n"); return(-1);} // See if it replies - want ACK n = read(fd,&resp,1); printf("%i bytes available, read: %i\n",n,resp); if (resp != ack) printf("Expecting , got %i\n",resp); // Send EOT n = write(fd,&eot,1); if (n != 1) {printf("Failed to write .\n"); return(-1);} // Close the port n = close(fd); if (n != 0) { printf("failed to close port: close returned %d\n",n); return(-1); } return 0; } void set_port(int fd) { struct termios options; int status; // Get current settings for the port tcgetattr(fd, &options); // Set baud rate = 1200 cfsetispeed(&options, B1200); cfsetospeed(&options, B1200); // Set 7 bits, even parity options.c_cflag |= PARENB; options.c_cflag &= ~PARODD; options.c_cflag &= ~CSTOPB; options.c_cflag &= ~CSIZE; options.c_cflag |= CS7; // Raw input options.c_lflag &= ~(ICANON | ECHO | ECHOE | ISIG); // Check and strip parity bit. options.c_iflag |= (INPCK | ISTRIP); // Disable software flow control options.c_iflag &= ~(IXON | IXOFF | IXANY); // Output to be raw options.c_oflag &= ~(OPOST | OLCUC | ONLCR | OCRNL ); // Set new options for port status = tcsetattr(fd, TCSANOW, &options); if (status < 0) { errx(1,"set_port(): failed."); } } int open_port(void) { int fd; fd = open("/dev/ttyU0", O_RDWR | O_NOCTTY | O_NDELAY); if (fd < 0) { errx(1,"open_port: Unable to open /dev/ttyS0."); } else { fcntl(fd, F_SETFL, 0); } return (fd); } dmesg from ALIX: OpenBSD 4.6 (GENERIC) #0: Thu Apr 8 14:17:46 BST 2010 r...@alix:/usr/src/sys/arch/i386/compile/GENERIC cpu0: Geode(TM) Integrated Processor by AMD PCS ("AuthenticAMD" 586-class) 499 MHz cpu0: FPU,DE,PSE,TSC,MSR,CX8,SEP,PGE,CMOV,CFLUSH,MMX real mem = 268009472 (255MB) avail mem = 250
Re: Serial port programming problems
On Sat, May 01, 2010 at 15:30:28 -0700, J.C. Roberts wrote: > > status = tcsetattr(fd, TCSANOW, &options); > > How does it behave if you use "TCSAFLUSH" rather than "TCSANOW" ? I made that substitution and added a #define DEBUG 1 The resulting binary sometimes fails to return and I then have to hit ctrl-c. I've tried it several times, and not seen any (repeating) pattern in terms of when it works and when it fails. Here's some typical output: # grep TCSA ex_obsd_flush.c status = tcsetattr(fd, TCSAFLUSH, &options); # cc -Wall -o ex_obsd_flush ex_obsd_flush.c # time ./ex_obsd_flush Port has been opened and set up 1 bytes available, read: 6 0m1.21s real 0m0.00s user 0m0.00s system # time ./ex_obsd_flush Port has been opened and set up ^C0m6.07s real 0m0.00s user 0m0.00s system # time ./ex_obsd_flush Port has been opened and set up ^C0m7.49s real 0m0.00s user 0m0.00s system # time ./ex_obsd_flush Port has been opened and set up 1 bytes available, read: 6 0m1.21s real 0m0.00s user 0m0.00s system Following this I had two more failures, one success, one failure and one success in that order. -- Neil
Re: Serial port programming problems
On Sun, May 2, 2010 at 5:33 PM, Neil O'Brien wrote: > On Sat, May 01, 2010 at 15:30:28 -0700, J.C. Roberts wrote: >> > status = tcsetattr(fd, TCSANOW, &options); >> >> How does it behave if you use "TCSAFLUSH" rather than "TCSANOW" ? > > I made that substitution and added a > #define DEBUG 1 > > The resulting binary sometimes fails to return and I then have to > hit ctrl-c. I've tried it several times, and not seen any (repeating) > pattern in terms of when it works and when it fails. > > Here's some typical output: > > # grep TCSA ex_obsd_flush.c > status = tcsetattr(fd, TCSAFLUSH, &options); > # cc -Wall -o ex_obsd_flush ex_obsd_flush.c > # time ./ex_obsd_flush > Port has been opened and set up > 1 bytes available, read: 6 >0m1.21s real 0m0.00s user 0m0.00s system > # time ./ex_obsd_flush > Port has been opened and set up > ^C0m6.07s real 0m0.00s user 0m0.00s system > > # time ./ex_obsd_flush > Port has been opened and set up > ^C0m7.49s real 0m0.00s user 0m0.00s system > > # time ./ex_obsd_flush > Port has been opened and set up > 1 bytes available, read: 6 >0m1.21s real 0m0.00s user 0m0.00s system > > Following this I had two more failures, one success, one failure and > one success in that order. > > -- > Neil > > Hi, Is the system running Linux running on the same type of board as your OpenBSD system? If not, try running Linux on a system identical to the one you're using - see if it's the board causing the problem. Regards -- Aaron Mason - Programmer, open source addict I've taken my software vows - for beta or for worse
Re: Serial port programming problems
On Sun, May 02, 2010 at 20:56:36 +1000, Aaron Mason wrote: > On Sun, May 2, 2010 at 5:33 PM, Neil O'Brien wrote: > > On Sat, May 01, 2010 at 15:30:28 -0700, J.C. Roberts wrote: > >> > status = tcsetattr(fd, TCSANOW, &options); > >> > >> How does it behave if you use "TCSAFLUSH" rather than "TCSANOW" ? > > > > I made that substitution and added a > > #define DEBUG 1 > > > > The resulting binary sometimes fails to return and I then have to > > hit ctrl-c. I've tried it several times, and not seen any (repeating) > > pattern in terms of when it works and when it fails. > > > Is the system running Linux running on the same type of board as your > OpenBSD system? If not, try running Linux on a system identical to > the one you're using - see if it's the board causing the problem. > Hi Aaron, No, they are very different machines. I don't have a spare ALIX system that I can load Linux onto for testing. If I can't resolve it any other way I will switch my current machine to Linux; but having just got everything set up in OpenBSD, I'd like to avoid that if possible. I have however tried the code in my initial email on a different OpenBSD machine (a Dell Optiplex Pentium II, dmesg is at http://www.tigerturnings.dyndns.dk/dmesg-dell) In this case I used cua0, the onboard serial port, on an ns16550a controller. Running the example from my initial post, with #define DEBUG 1 added and /dev/ttyU0 replaced by /dev/cua00, I get sporadic failures: # time ./ex_obsd_now_cua0 ; sleep 3 ; time ./ex_obsd_now_cua0 Port has been opened and set up 1 bytes available, read: 6 0m0.12s real 0m0.00s user 0m0.00s system Port has been opened and set up ^C0m7.00s real 0m0.00s user 0m0.00s system If I replace TCSANOW with TCSAFLUSH on this machine, it doesn't help, though the instrument does sometimes send 21 (NAK) which I've never seen it do when using TCSANOW. # time ./ex_obsd_flush_cua0 Port has been opened and set up 1 bytes available, read: 6 0m0.14s real 0m0.00s user 0m0.00s system # time ./ex_obsd_flush_cua0 Port has been opened and set up ^C0m3.97s real 0m0.00s user 0m0.01s system # time ./ex_obsd_flush_cua0 Port has been opened and set up ^C0m3.65s real 0m0.00s user 0m0.00s system # time ./ex_obsd_flush_cua0 Port has been opened and set up 1 bytes available, read: 21 Expecting , got 21 0m0.23s real 0m0.00s user 0m0.00s system I think that this should rule out any problems with the USB-RS232 converter or its drivers, and with the USB controller on the ALIX system. I'd be grateful for any further ideas or suggestions. -- Neil
Re: Serial port programming problems
On Sun, 2 May 2010 16:45:44 +0100 "Neil O'Brien" wrote: > On Sun, May 02, 2010 at 20:56:36 +1000, Aaron Mason wrote: > > On Sun, May 2, 2010 at 5:33 PM, Neil O'Brien > > wrote: > > > On Sat, May 01, 2010 at 15:30:28 -0700, J.C. Roberts wrote: > > >> > status = tcsetattr(fd, TCSANOW, &options); > > >> > > >> How does it behave if you use "TCSAFLUSH" rather than "TCSANOW" ? > > > > > > I made that substitution and added a > > > #define DEBUG 1 > > > > > > The resulting binary sometimes fails to return and I then have to > > > hit ctrl-c. I've tried it several times, and not seen any > > > (repeating) pattern in terms of when it works and when it fails. > > > > > Is the system running Linux running on the same type of board as > > your OpenBSD system? If not, try running Linux on a system > > identical to the one you're using - see if it's the board causing > > the problem. > > > > Hi Aaron, > > No, they are very different machines. I don't have a spare ALIX > system that I can load Linux onto for testing. If I can't resolve it > any other way I will switch my current machine to Linux; but having > just got everything set up in OpenBSD, I'd like to avoid that if > possible. > > I have however tried the code in my initial email on a different > OpenBSD machine (a Dell Optiplex Pentium II, dmesg is at > http://www.tigerturnings.dyndns.dk/dmesg-dell) > > In this case I used cua0, the onboard serial port, on an ns16550a > controller. Running the example from my initial post, with #define > DEBUG 1 added and /dev/ttyU0 replaced by /dev/cua00, I get sporadic > failures: > > # time ./ex_obsd_now_cua0 ; sleep 3 ; time ./ex_obsd_now_cua0 > Port has been opened and set up > 1 bytes available, read: 6 > 0m0.12s real 0m0.00s user 0m0.00s system > Port has been opened and set up > ^C0m7.00s real 0m0.00s user 0m0.00s system > > If I replace TCSANOW with TCSAFLUSH on this machine, it doesn't help, > though the instrument does sometimes send 21 (NAK) which I've never > seen it do when using TCSANOW. > > # time ./ex_obsd_flush_cua0 > Port has been opened and set up > 1 bytes available, read: 6 > 0m0.14s real 0m0.00s user 0m0.00s system > # time ./ex_obsd_flush_cua0 > Port has been opened and set up > ^C0m3.97s real 0m0.00s user 0m0.01s system > > # time ./ex_obsd_flush_cua0 > Port has been opened and set up > ^C0m3.65s real 0m0.00s user 0m0.00s system > > # time ./ex_obsd_flush_cua0 > Port has been opened and set up > 1 bytes available, read: 21 > Expecting , got 21 > 0m0.23s real 0m0.00s user 0m0.00s system > > I think that this should rule out any problems with the USB-RS232 > converter or its drivers, and with the USB controller on the ALIX > system. > > I'd be grateful for any further ideas or suggestions. > > -- Neil > I should be able to figure this out, but my C language serial port hacking skills are a bit rusty. These days, C is only used for time-critical portions of instrument automation for test/mfg environments, but for everything else, languages like perl, tcl or custom/proprietary LabView executables are used. Lots of instruments speak "skippy" (SCPI) otherwise known as Standard Commands for Programmable Instruments. Your *UNNAMED* instrument might speak a custom protocol as you claimed, or it might also speak "skippy" either natively or with some configuration. http://en.wikipedia.org/wiki/Standard_Commands_for_Programmable_Instrumentation If your instrument can speak "skippy" then your life gets a whole lot easier, so it's definitely worth checking. I'm not sure why you failed to post the make and model number details of your instrument, but they would be really helpful. As I mentioned off list, TCSAFLUSH is often required, but more than that, it's just good practice to flush the buffers to make sure you're starting in a known state (e.g. no crap in the buffers). Both your C code, and not understanding the behavior and output you posted seems to indicate you followed one of the countless (poorly written) how-to pages or books like: http://www.easysw.com/~mike/serial/serial.html The above has some good spots but it also has plenty of errors and unfortunately, it is highly ranked by search engines. In your code you seem to have inherited one of the mistakes, namely you think read(2) gives you "bytes available" rather than "bytes read" as stated (twice) in the read(2) man page. $ man 2 read ... Upon successful completion, read(), readv(), pread(), and preadv() return the number of bytes actually read and placed in the buffer. The system guarantees to read the number of bytes requested if the descriptor references a normal file that has that many bytes left before the end-of- file, but in no other case. ... The other problem is understanding what is happening. Unless you specifically configured the descriptor to return immediately, your read(2) call will sleep
Re: Serial port programming problems
On Sun, 2 May 2010 12:01:59 -0700 "J.C. Roberts" wrote: > The other problem is understanding what is happening. Unless you > specifically configured the descriptor to return immediately, your > read(2) call will sleep until it gets the requested number of bytes > from the descriptor, until an interval timer expires, or until an > error occurs. Since the instrument is not sending you any data, it > sleeps until you lose your patience and hit CTL-C to end the program. > --This should not be a surprise. ;) I should have referenced your code so the above makes sense. >> int open_port(void) >> { >> int fd; >> >> fd = open("/dev/ttyU0", O_RDWR | O_NOCTTY | O_NDELAY); >> if (fd < 0) >> { >> errx(1,"open_port: Unable to open /dev/ttyS0."); >> } >> else >> { >> fcntl(fd, F_SETFL, 0); >> } >> return (fd); >> } Setting O_NDELAY (or better said O_NONBLOCK) on open gets wiped out when you run "fcntl(fd, F_SETFL, 0)" I have no idea why you're doing that (probably just mistyped example code), but if you insist, then you'd actually want: fcntl(fd, F_SETFL, FNDELAY); or fcntl(fd, F_SETFL, FNONBLOCK); Hence the reason why you're actually blocking on read. -- The OpenBSD Journal - http://www.undeadly.org