I have a linux platform* that is connected as a usb device to an automotive device which acts as the USB host. The two devices should communicate over CDC/NCM, but the linux platform is not recognised by the automotive device and therefore the connection is not established. Surprisingly a connection to my computer is established correctly.
I now need to create a trace of that USB connection in order to check if there is an error in the USB handshake that can't be handled by the automotive device. As I cannot access the USB host, I need to create the trace from the gadget side.
I tried using usbmon and tcpdump, but this seems to work only for USB controllers configured as hosts on the tracing platform, not for ones configured as devices.
How can I configure usbmon to work also on devices?
If that is not possible are there any other possibilities to achieve this? (preferrably without hacking any drivers...)
Or do I have to use a Hardware USB sniffer?
BTW, all required modules (esp. g_ncm) are correctly loaded.
Thank you for your help!
stefan
*custom distribution on a freescale iMX6 processor (ARM), Kernel Version 3.0.35
Related
What I want to do:
An AI program on a host machine, reading inputs from a camera sensing the screen of the target machine and outputting controls to the target machine via USB connection--programming the host machine's USB host as a USB peripheral connected to the target machine.
What I want to do step by step: (is it possible to implement the steps below?)
Have a host machine A and a target machine B.
Connect A and B with a USB 3.0 Type-A male-male cable.
The USB connection shows up as an HID keyboard device on B.
Write code to simulate key presses on A that sends to B.
(Eg. calling press('F') on a program running on A would type F to B's input)
It shouldn't require any program installed on B.
What I already searched:
USB 3.0 Host to host connection is possible:
https://superuser.com/questions/795053/how-do-i-connect-two-computers-using-usb-3-0
USB 2.0 Host to host connection is impossible:
https://superuser.com/questions/99274/how-to-connect-two-computers-with-usb
Similar questions asked without the assumption that USB 3.0 Host to Host connection is possible:
https://superuser.com/questions/1128365/simulate-usb-keyboard-from-machine
Setting up a computer to act as an HID device connected to another computer via ps/2,usb or another wired connection
https://superuser.com/questions/507921/computer-to-act-as-keyboard?rq=1
Suggestions in ascending order of feasibility:
USB Gadgets
You are using linux, so the default way would be to create/configure/load a gadget driver. Have a look at this tutorial, though for a raspberry, should work on your PC too. However, I could not find any information regarding the use of USB3 - the tutorial assumes your host is using one of it's OTG ports, which your PC most likely does not have. So whether this works with your USB3.1 Type-A-to-Type-A connection you'll need to test.
USBIP
The idea of sharing USB devices (not just keyboards) is not really new. With USBIP you can "export" any local USB device to the network, and your client will need the client-side USBIP driver to access the keyboard.
Dont bother with USB at all, just use Ethernet
I'd simply write two userland scripts/programs that send/receive+execute the keystrokes. Very easy to implement, you're probably familiar with python anyway.
If you absolutely cant have software installed on the client-PC and your Type-C-to-Type-C connection doesnt support USB Gadgets, there's another way. It basically involves the use of two USB-to-serial adapters (~15$) and a serial cable. While this wont be enumerated as a keyboard, but rather as serial port, it's the lowest-effort solution to transfer data without additional software on the client. Both computers will just do file I/O. If your computers still have COM-ports, you can even omit the serial converters!
I would like to implement a USB device driver in user space by means of libusb. I'm using a Linux machine supporting a USB OTG controller which is switched to device mode. The USB host is another machine which needs to communicate with my Linux machine by means of a USB vendor specific interface with a bulk in/out interface.
I would like to know if it is possible to use libusb to communicate with the USB host on the other side. Or, if libusb could be used only for host side functionalities.
If libusb can't be used is there any other way to implement the device driver in user space?
Thank you.
As of my knowledge this is not possible. (Vanilla) libusb is only for host usage because the whole process of how to use libusb can only provide this mode.
You can find an answer in the FAQ of the old libusb (before libusbx, the abandoning of the old libusb and the rename of libusbx to the new libusb):
libusb FAQ
Also there is a question on SO to this topic with some suggestions:
How to communicate with the USB Host from a Linux USB Client
I am on debian and:
I have a USB controller hooked up to a USB port on my PC (Device 1).
I have a male to male USB cord hooked up to another port on the PC that connects to Device 2. (it is a "bridging" usb cord, and has the chip for it)
I want to make them connect to each other as if they were one cord, so neither device knows that there is a computer in the middle.
This would be called a 'Coupler', except that I am using a PC as a coupler.
Here is a (really bad) diagram I made:
What I have done:
I have been able to connect the two devices independently of each other and sniff the results for when they fail to connect. The devices don't send a large volume of data back and forth.
Maybe there is some kind of command tool that I could use, for example (psudocode):
$ couple-usb-ports PORT1 PORT2
You're trying to reinvent the wheel here.
You might consider looking at this link instead.
http://dan3lmi.blogspot.com/2012/10/sniffing-usb-traffic-different.html
Specifically this.
Windows: You cannot directly capture raw USB traffic on Windows with Wireshark/WinPcap, but it is possible to capture and debug USB traffic on a virtual Windows machine under Oracle Virtual Box.
You cannot use a simple PC as transparent USB sniffer without extra (expensive) hardware. An USB bus has always one host (and one or more devices), and the PC can only be the host. This is a hardware limitation.
But you can capture USB data in a Windows machine using Wireshark and USBPcap, eliminating the need for the middle box in most cases.
As this post is tagged Linux, I suppose the controller PC is a Linux machine. Instead of connecting USB ports with a male-male connector, which is all kinds of bad (you are connecting the 5V lines of both machine with each other!), just run Wireshark in the controller PC.
There might be a little work to be done previously, as you have to enable Wireshark for USB monitoring (Particularly in Debian, this is disabled by default), and you might have to install a small driver to enable the monitoring. Have a look at this page for more information.
Once you get it working, Wireshark is an excellent tool for this!
I'm looking to attach some USB devices to my embedded Linux board.
It is an TI-ARM processor running embedded Linux, but I guess it could be any embedded Linux board.
If I purchase an USB device which has Linux support/driver, can this driver (generally) be re-compiled to work with the ARM architecture? (Instead of Windows ect.).
Yes, USB drivers can generally be expected to compile for other architectures other than x86. Of course this presumes that your board does have a host USB port. There are a few boards that have only USB device ports, and many SoCs have both USB host & device ports.
But successfully compiling the (USB) driver may only be part of the task.
Some (USB) devices may require additional packages of libraries and other drivers for interfacing to application programs. For instance a USB digital TV tuner requires numerous packages (V4L, ALSA, I2C driver, userland firmware loading) to actually work.
Clarification
These additional dependencies that you may have to build are not because of USB.
The dependencies are related to the type of device.
An Ethernet interface, whether integrated into the SoC or offboard using USB, would be easily configured for full support in the kernel (e.g. protocol stack) and userland (e.g. Busybox has ifconfig, ping and routing apps).
A PCI TV tuner would have the same dependencies as the USB tuner. But the embedded environment typically means that you don't have any/most of these multimedia dependencies already built/installed.
I have a Linux virtual machine running on VMWare Fusion (on Mac OS X) that I intend to use as a development environment for an embedded system. Would it be possible for my Linux VM to talk to my embedded system's serial port using a USB-to-serial adapter? Any recommendations for what sort of adapter I should get?
There are two ways to do it:
Your host-OS supports your USB<->serial converter (very likely). If so you can just allow your VM to talk to the serial port. If so the VM will see a standard serial port and everything will be fine.
Your guest-os in the VM supports the USB<->serial converter, AND your host-OS allows raw USB forwarding.
All in all the chances are good that it works..
Btw: there are good and bad USB<->Serial converters. If you find out that the serial connection seems to work (everything detects/works as expected for a couple of seconds), but you can't get a reliable connection for a longer time, then it's very possible that the usb<->serial dongle sucks....
You get what you pay for... When buying these things I'd check comp.arch.embedded and ask which usb dongles are known to just work, and which not. (My recommendation is to stick with products from Assmann. You can order them at digikey).
I've found USB serial adaptors to be a bit hit & miss with embedded work. One thing to be aware of is that the buffering tends to work differently from "real" serial ports, and latency of characters through the system can be quite variable. Some embedded development systems (think bootloaders, cheap JTAG probes etc) can be quite sensitive to this and will give timeouts and so on.
Note this doesn't only apply to USB serial adaptors, I've had similar problems with high end multi-port serial cards, but usually with those you can tweak the FIFO / IRQ settings to get something working.
I have experienced that a USB to serial adapter with a FTDI chipset and drivers is more reliable and compatible with more devices than the Prolific chipset
Depends upon the VM software, but VMWare Fusion does support USB devices. The question becomes, does your IDE support talking to a USB device instead of an old-fashioned serial port? With Linux, probably yes.
I had no problems whit serial adapters from ATEN.
USB serial adapter is USB standard device (just like mass storage) that mean that any USB compliant adapter should work.
pl2303: I have found this device to be very reliable and are often in the generic and cheap USB to RS232 adapters. I've seen expensive adapters fail and my generic adapter from geeks.com work great.
I just picked up a USB 1.1 - RS232 adapter (Digitus DA-70119) from WeirdStuff for 10 bucks. I plugged it into my Mac mini and VMWare Fusion showed me this.
Once I clicked on the USB icon, my Ubuntu 9.10 VM had no trouble seeing it
$ lsusb
Bus 002 Device 004: ID 067b:2303 Prolific Technology, Inc. PL2303 Serial Port
$ dmesg
usb 2-1: new full speed USB device using uhci_hcd and address 4
usb 2-1: configuration #1 chosen from 1 choice
pl2303 2-1:1.0: pl2303 converter detected
usb 2-1: pl2303 converter now attached to ttyUSB0
I can now use /dev/ttyUSB0 in my Linux VM to talk to my target system.