11 ms·

A lot of hardware still uses serial, and not just industrial stuff. Everything from sewing machines to remote controlled cameras. If you work on embedded devic

A lot of hardware still uses serial, and not just industrial stuff. Everything from sewing machines to remote controlled cameras. If you work on embedded devices you will still encounter serial/RS-232 all the time. Often through USB-to-serial chips, which only adds to the challenge because they are mostly unreliable crap. Then there are about 30 parameters to configure on a TTY. About half do absolutely nothing, a quarter completely breaks the signal giving you silence or line noise, the final quarter only subtly breaks the signal, occasionally corrupting your data. Still, there is nothing like injecting a bootloader directly to RAM over JTAG, running that to get serial and upload a better bootloader, writing it to flash and finally getting ethernet and TCP/IP up.

gcp 10y ago
Arduino got so popular because it addressed those exact problems: realiable USB to serial built in, foolproof bootloader. Still, there is nothing like injecting a bootloader directly to RAM over JTAG, running that to get serial and upload a better bootloader, writing it to flash and finally getting ethernet and TCP/IP up. I'm happy to have gotten mostly rid of this. Gone are the days of choosing motherboards based on the LPT support, and praying that the JTAG drivers would work on an OS upgrade.
- kosma 10y ago
  > I'm happy to have gotten mostly rid of this. Gone are the days of choosing motherboards based on the LPT support, and praying that the JTAG drivers would work on an OS upgrade. It's still there, and not going anywhere - the only thing that's past is LPT. Last time I used a Linux-grade Atmel SoC, it had a USB-CDC interface but the chain was still the same: boot from mask ROM, get minimal USB bootloader, load a bootstrap binary to SRAM, use that to initialize external DRAM, then load a flashing applet to DRAM, run it, use that to burn u-boot to flash, and then fire up u-boot's Ethernet & TFTP client to start a kernel from an external server and mount rootfs over NFS. Considering the amount of magic, it worked amazingly well. The whole shebang was packaged into a zip file with a single BAT to double-click and let it do the magic. As for COM and LPT - FTDI and J-Link changed the embedded landscape forever, and thanks for that.
- stuckagain 10y ago
  Arduino got so popular because it does everything the Atmel STK does at a fraction of the price.
- kbob 10y ago
  Get yourself a Black Magic Probe. It's JTAG/SWD on the device end, USB GDB server on the host end. https://hackaday.com/2016/12/02/black-magic-probe-the-best-arm-jtag-debugger/ https://hackaday.com/2016/12/02/black-magic-probe-the-best-a... disclaimer: I'm friends with the BMP manufacturer.
osullivj 10y ago
Yes! I'm currently working on a Windows based building access system dating back to the early 90s. Most of the codebase in MFC C++. The hardware/firmware that runs the door controls used to talk RS232 to the Windows boxes, but now use TCP/IP. STX is used as the start char for a msg. The system is used by some very prominent govt and biz orgs.
- kosma 10y ago
  One pleasant upside of such systems is that the protocols are usually so simple you can reverse engineer them in a couple evenings. I can't count the times I ended up deriving the serial protocol from some undocumented, baroque source code (or sniffer dumps) and writing a quick & dirty implementation in a few dozen lines of Python.
  - osullivj 10y ago
    That's exactly what I've done :) The checksum algorithm that suffixes each msg with a 2 byte check took a little bit of figuring out. I'm using tornado.tcpserver.TCPServer and tornado.iostream. Another plus is generating logs of the msg flow; useful for diagnostics, and the after the fact specs you often need with crufty old systems.
drvdevd 10y ago
Isn't there something like, just one company in the world that makes 95% of all USB to Serial chips?
- kosma 10y ago
  It used to be "just" FTDI... but then other manufacturers and endless copycats jumped in, which resulted in a XKCD #927 situation. As if we didn't already have a USB-CDC standard... (For an interesting side story, search for "ftdigate". At some point FTDI decided they're fed up with copycats and released a new driver pack (through automatic Windows Update) that bricked counterfeit chips. This led to a lot of angry people and number of amusing situations, including someone jokingly submitting a patch to Linux to do the same.)
  - dfox 10y ago
    FTDI is not and never was especially big in the USB-to-Serial market, and their chips tended to implement CDC in variously broken ways not to mention that their windows drivers are major PITA. What they are big in is market for "just add USB to this custom device" because they supply mostly NDA-free documentation since forever and produce various ready-made modules.
    - kosma 10y ago
      Speaking out of curiosity, can you name some companies that provide NDA-laden USB-serial bridges? I'm painfully aware that it's hard to even know such products exist, especially when your production volume barely runs into hundreds or thousands and you're not established in the industry... and from my limited interactions with even the smaller players like SIMCom or SiRF, it takes an NDA just to talk to someone who can get you to sign a second NDA so you can see datasheets. Even if you have tons of money, giving it to such company in return for their product can be a challenge.
    - ajford 10y ago
      They also put the legwork into getting their drivers into nearly everything, which makes them great to use in a custom or small batch product that has users that aren't technically oriented or wont be installed by a knowledgeable installer. If you already have a client side install (like a software package) that can bundle the drivers for you, it's not a big deal. And with the emergence of USB-CDC, it's not nearly as necessary as it used to be, since most modern OSes support that now.
heywire 10y ago
Definitely. Just this week actually I'm writing an interface to a _new_ RS232 device for work.
xbryanx 10y ago
Just about every professional display or projector has a RS-232 control interface. In arenas, museums, airports, restaurants, any place you see higher end digital signage, you'll likely see a RS-232 serial cable connected to the display device for control and maintenance.
SixSigma 10y ago
On Plan9 you can run TCP/IP over RS232 and RS232 over TCP/IP with ~zero effort too. So your sensors can talk across the internet.
- i336_ 10y ago
  I wish Plan9 were practically usable (good graphics support, modern chipset support, etc); it's the lack of those types of critical features that limit its widespread use at the end of the day. From `man socat', which runs on Linux: (socat PTY,link=$HOME/dev/vmodem0,raw,echo=0,wait-slave EXEC:'"ssh modemserver.us.org socat - /dev/ttyS0,non-block,raw, echo=0"') generates a pseudo terminal device (PTY) on the client that can be reached under the symbolic link $HOME/dev/vmodem0. An appli- cation that expects a serial line or modem can be configured to use $HOME/dev/vmodem0; its traffic will be directed to a modem- server via ssh where another socat instance links it with /dev/ttyS0.
  - SixSigma 10y ago
    import /n/kremvax/net /net.alt now kremvax' network stack is bound over my own no need for nat
    - i336_ 10y ago
      Okay that's really impressive. By "bound over", do you mean tunneled, or "made available alongside"?
      - SixSigma 10y ago
        In Plan9 the file system is per process so I can choose to bind kremvax' network over mine in this shell window, then pentvax' network over min in another. I think I'm right in sayig that if one has /net and /net.alt then /net is asked first and then /net.alt I'm a bit rusty on the details To make a connection a process open's its /net/tcp/ctrl file, writes a connection string, gets back a response. If that response is a number then it opens /net/tcp/1/ctrl and /net/tcp/1/data and read / writes data to the data and sees out of stream messages on ctrl. To close the connection one closes /net/tcp/1/ctrl Everything is done via the 9p protocol. So if I can write code on my Arduino that understands 9p and a way to send/rec that data (e.g. serial but even SP1 or JTAG) to a Plan9 machine locally, then the Arduino itself could just open a tcp connection on kremvax. Once you start layering these things on top of each other it gets a bit mind blowing and you get drunk on power. Then when you are forced back Linux or Windows you realise how dumb they are even with their fancy application software.
digler999 10y ago
isn't one reason that RS-232 is an open protocol but USB requires licensing ? I thought it cost money to get an official VID/PID from the USB organization ?
CaptSpify 10y ago
I used to work on x-ray equipment that still used serial/rs-232 ~4 years ago. I'm sure they are still around. I had to set up multiboot laptops that ran win 95/98/xp, because we couldn't virtualize the serial connection properly. And yes, the USB-to-serial rarely worked either. Man that job was frustrating, but it sure was a lot of fun!
avocad 10y ago
It wouldn't surprise me if RS-232 will still be around long after USB falls out of use. Does anyone know the reasons why USB was not made backward compatible with RS-232? It would only take a very short negotiation to determine if both endpoints support USB.
- esmi 10y ago
  I don't know but my guess is to add compatibility for RS232 voltage levels would add significant cost and complexity to the USB PHY, the thing that actually drives and measures voltages on the wire. USB was specifically designed to have a simple and cheap PHY so that it could be used in very cheap peripherals. To expand a bit, the typical fully compliant RS232 setup has a special level translation IC (for example the MAX232) to deal with the huge voltage range and convert it to something more low voltage digital logic friendly. And that IC usually requires power supply levels that the typical cheap USB device would not have, which means it would have to add more circuits to generate them, adding even more cost to the hardware. And the USB committee probably figured to be not fully compliant with RS232 would just confuse people, and damage hardware, so it was better to be not compliant at all.
  - i336_ 10y ago
    It would have been really awesome if USB was compatible with TTL-level (5v, maybe 3.3v) "serial". This nonstandard variant of definitely-not-RS232 is everywhere. But... > And the USB committee probably figured to be not fully compliant with RS232 would just confuse people, and damage hardware, so it was better to be not compliant at all. ...you are sadly right.
    - throwawayish 10y ago
      USB controllers have to support 3.3 V anyway (on the D+/D- pair, not on the SS pairs), for supporting USB 1.x modes.
altnewman 10y ago
Serial is so easy to program for, and really has first class support on the most common/classic SOCs. USB is kind of tough TBH.

gcp 10y ago

Arduino got so popular because it addressed those exact problems: realiable USB to serial built in, foolproof bootloader. Still, there is nothing like injecting a bootloader directly to RAM over JTAG, running that to get serial and upload a better bootloader, writing it to flash and finally getting ethernet and TCP/IP up. I'm happy to have gotten mostly rid of this. Gone are the days of choosing motherboards based on the LPT support, and praying that the JTAG drivers would work on an OS upgrade.

kosma 10y ago
> I'm happy to have gotten mostly rid of this. Gone are the days of choosing motherboards based on the LPT support, and praying that the JTAG drivers would work on an OS upgrade. It's still there, and not going anywhere - the only thing that's past is LPT. Last time I used a Linux-grade Atmel SoC, it had a USB-CDC interface but the chain was still the same: boot from mask ROM, get minimal USB bootloader, load a bootstrap binary to SRAM, use that to initialize external DRAM, then load a flashing applet to DRAM, run it, use that to burn u-boot to flash, and then fire up u-boot's Ethernet & TFTP client to start a kernel from an external server and mount rootfs over NFS. Considering the amount of magic, it worked amazingly well. The whole shebang was packaged into a zip file with a single BAT to double-click and let it do the magic. As for COM and LPT - FTDI and J-Link changed the embedded landscape forever, and thanks for that.
stuckagain 10y ago
Arduino got so popular because it does everything the Atmel STK does at a fraction of the price.
kbob 10y ago
Get yourself a Black Magic Probe. It's JTAG/SWD on the device end, USB GDB server on the host end. https://hackaday.com/2016/12/02/black-magic-probe-the-best-arm-jtag-debugger/ https://hackaday.com/2016/12/02/black-magic-probe-the-best-a... disclaimer: I'm friends with the BMP manufacturer.

osullivj 10y ago

Yes! I'm currently working on a Windows based building access system dating back to the early 90s. Most of the codebase in MFC C++. The hardware/firmware that runs the door controls used to talk RS232 to the Windows boxes, but now use TCP/IP. STX is used as the start char for a msg. The system is used by some very prominent govt and biz orgs.

kosma 10y ago
One pleasant upside of such systems is that the protocols are usually so simple you can reverse engineer them in a couple evenings. I can't count the times I ended up deriving the serial protocol from some undocumented, baroque source code (or sniffer dumps) and writing a quick & dirty implementation in a few dozen lines of Python.
- osullivj 10y ago
  That's exactly what I've done :) The checksum algorithm that suffixes each msg with a 2 byte check took a little bit of figuring out. I'm using tornado.tcpserver.TCPServer and tornado.iostream. Another plus is generating logs of the msg flow; useful for diagnostics, and the after the fact specs you often need with crufty old systems.

drvdevd 10y ago

Isn't there something like, just one company in the world that makes 95% of all USB to Serial chips?

kosma 10y ago
It used to be "just" FTDI... but then other manufacturers and endless copycats jumped in, which resulted in a XKCD #927 situation. As if we didn't already have a USB-CDC standard... (For an interesting side story, search for "ftdigate". At some point FTDI decided they're fed up with copycats and released a new driver pack (through automatic Windows Update) that bricked counterfeit chips. This led to a lot of angry people and number of amusing situations, including someone jokingly submitting a patch to Linux to do the same.)
- dfox 10y ago
  FTDI is not and never was especially big in the USB-to-Serial market, and their chips tended to implement CDC in variously broken ways not to mention that their windows drivers are major PITA. What they are big in is market for "just add USB to this custom device" because they supply mostly NDA-free documentation since forever and produce various ready-made modules.
  - kosma 10y ago
    Speaking out of curiosity, can you name some companies that provide NDA-laden USB-serial bridges? I'm painfully aware that it's hard to even know such products exist, especially when your production volume barely runs into hundreds or thousands and you're not established in the industry... and from my limited interactions with even the smaller players like SIMCom or SiRF, it takes an NDA just to talk to someone who can get you to sign a second NDA so you can see datasheets. Even if you have tons of money, giving it to such company in return for their product can be a challenge.
  - ajford 10y ago
    They also put the legwork into getting their drivers into nearly everything, which makes them great to use in a custom or small batch product that has users that aren't technically oriented or wont be installed by a knowledgeable installer. If you already have a client side install (like a software package) that can bundle the drivers for you, it's not a big deal. And with the emergence of USB-CDC, it's not nearly as necessary as it used to be, since most modern OSes support that now.

heywire 10y ago

Definitely. Just this week actually I'm writing an interface to a _new_ RS232 device for work.

xbryanx 10y ago

Just about every professional display or projector has a RS-232 control interface. In arenas, museums, airports, restaurants, any place you see higher end digital signage, you'll likely see a RS-232 serial cable connected to the display device for control and maintenance.

SixSigma 10y ago

On Plan9 you can run TCP/IP over RS232 and RS232 over TCP/IP with ~zero effort too. So your sensors can talk across the internet.

i336_ 10y ago
I wish Plan9 were practically usable (good graphics support, modern chipset support, etc); it's the lack of those types of critical features that limit its widespread use at the end of the day. From `man socat', which runs on Linux: (socat PTY,link=$HOME/dev/vmodem0,raw,echo=0,wait-slave EXEC:'"ssh modemserver.us.org socat - /dev/ttyS0,non-block,raw, echo=0"') generates a pseudo terminal device (PTY) on the client that can be reached under the symbolic link $HOME/dev/vmodem0. An appli- cation that expects a serial line or modem can be configured to use $HOME/dev/vmodem0; its traffic will be directed to a modem- server via ssh where another socat instance links it with /dev/ttyS0.
- SixSigma 10y ago
  import /n/kremvax/net /net.alt now kremvax' network stack is bound over my own no need for nat
  - i336_ 10y ago
    Okay that's really impressive. By "bound over", do you mean tunneled, or "made available alongside"?
    - SixSigma 10y ago
      In Plan9 the file system is per process so I can choose to bind kremvax' network over mine in this shell window, then pentvax' network over min in another. I think I'm right in sayig that if one has /net and /net.alt then /net is asked first and then /net.alt I'm a bit rusty on the details To make a connection a process open's its /net/tcp/ctrl file, writes a connection string, gets back a response. If that response is a number then it opens /net/tcp/1/ctrl and /net/tcp/1/data and read / writes data to the data and sees out of stream messages on ctrl. To close the connection one closes /net/tcp/1/ctrl Everything is done via the 9p protocol. So if I can write code on my Arduino that understands 9p and a way to send/rec that data (e.g. serial but even SP1 or JTAG) to a Plan9 machine locally, then the Arduino itself could just open a tcp connection on kremvax. Once you start layering these things on top of each other it gets a bit mind blowing and you get drunk on power. Then when you are forced back Linux or Windows you realise how dumb they are even with their fancy application software.

digler999 10y ago

isn't one reason that RS-232 is an open protocol but USB requires licensing ? I thought it cost money to get an official VID/PID from the USB organization ?

CaptSpify 10y ago

I used to work on x-ray equipment that still used serial/rs-232 ~4 years ago. I'm sure they are still around. I had to set up multiboot laptops that ran win 95/98/xp, because we couldn't virtualize the serial connection properly. And yes, the USB-to-serial rarely worked either. Man that job was frustrating, but it sure was a lot of fun!

avocad 10y ago

It wouldn't surprise me if RS-232 will still be around long after USB falls out of use. Does anyone know the reasons why USB was not made backward compatible with RS-232? It would only take a very short negotiation to determine if both endpoints support USB.

esmi 10y ago
I don't know but my guess is to add compatibility for RS232 voltage levels would add significant cost and complexity to the USB PHY, the thing that actually drives and measures voltages on the wire. USB was specifically designed to have a simple and cheap PHY so that it could be used in very cheap peripherals. To expand a bit, the typical fully compliant RS232 setup has a special level translation IC (for example the MAX232) to deal with the huge voltage range and convert it to something more low voltage digital logic friendly. And that IC usually requires power supply levels that the typical cheap USB device would not have, which means it would have to add more circuits to generate them, adding even more cost to the hardware. And the USB committee probably figured to be not fully compliant with RS232 would just confuse people, and damage hardware, so it was better to be not compliant at all.
- i336_ 10y ago
  It would have been really awesome if USB was compatible with TTL-level (5v, maybe 3.3v) "serial". This nonstandard variant of definitely-not-RS232 is everywhere. But... > And the USB committee probably figured to be not fully compliant with RS232 would just confuse people, and damage hardware, so it was better to be not compliant at all. ...you are sadly right.
  - throwawayish 10y ago
    USB controllers have to support 3.3 V anyway (on the D+/D- pair, not on the SS pairs), for supporting USB 1.x modes.

altnewman 10y ago

Serial is so easy to program for, and really has first class support on the most common/classic SOCs. USB is kind of tough TBH.