How to find a bug with lost network messages

[Arthur Pichlkostner]

I just know that netif_rx() should be updated to netif_rx_ni() for newer kernels.
Without the change I had NOHZ errors in the log, the same change was done in SLCAN.
Maybe this is the origin of your problem.

You can try our fork on https://github.com/tjohann/sllin which includes many improvents and fixes compared to the original driver from 2013.

On Tue, Feb 02, 2016 at 10:09:20AM +0100, Sandro Stiller wrote:
> Hello,
> 
> I'm struggeling with a network driver (sllin[1]) which is not in the 
> official kernel.
> It has a lot in common with the slcan driver but is used for LIN networks.
> The problem is, that sometimes messages sent to the network layer via 
> netif_rx() don't arrive in all listening programs.
> 
> This is how the driver works:
> 1. The application sends CAN messages to the network interface
> 2. The driver forwards it to the UART (tty)
> 3. The UART receives the same message (single-wire connection, RX and TX 
> connected) and sends it back to the network layer
> 4. The sending application receives the previously sent message and can 
> check for transmission errors and appended LIN slave replies.
> 
> Sometimes the last point (4.) does not work after 10 - 40 seconds of 
> transmission.
> The application does not receive the message using a blocking read() on 
> the socket, but other processes receive it (running candump on the 
> interface). netif_rx() always returns 0.
> 
> If more programs are listening (running multiple instances of candump), 
> the problem appears less often or never.
> On my PC there is no problem, it occures on ARM only.
> I'm using kernel 4.1.
> 
> Can you give me a hint where to search for the cause of this behaviour?
> 
> Thank you very much.
> 
> Sandro
> 
> 
> [1]: https://github.com/sstiller/sllin/tree/master/sllin
> 
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies