SK-AM62A-LP: Sometimes the rpmsg driver on Linux side can't get rpmsg_driver->callback function be called after Linux rpmsg_send() a msg to the internal MCU.

Hi Nick,

    Thanks for your fast reply. I will try your suggestion a little later. But I have 2 questions more:

1. Sometimes, I also met rpmsg timeout when Linux side tried to write a MCU register by rpmsg. (Not many data need be transfered), for this case, what is possible reason to rpmsg timeout?

2. Would you please help to provide the document or sample code about how to create and use a shared memory between Linux main CPU and R5 MCU？

Thanks and Regards,

David

Hello David,

Just to verify: Are you writing your own IPC here, or are you using driver provided by TI that is handling the IPC? e.g., are you using a TI CSI driver? (I don't know anything about the CSI usecase, so please forgive my ignorance about the usecase and which company is writing which code)

1.

Linux does not use RPMsg to write to the MCU registers. Linux RPMsg writes data to a VIRTIO buffer that is in a shared memory region in DDR, and then the remote core is notified to go read that RPMsg buffer.

So to double-check, are you writing directly to the MCU registers? If so, how are you doing that write, and what terminal output do you see?

Or are you just seeing timeouts when doing RPMsg at a "slower" pace? If so, what is the rate that RPMsg messages are sent in this "slower" pace?

2.

For an example of a shared memory region, please refer to the zerocopy example:
https://git.ti.com/cgit/rpmsg/rpmsg_char_zerocopy/ 

Note that there are many possible ways to implement shared memory - this is just one method. Please also note that dma-buf-phys is the Linux driver we use for the example on kernel 6.1, but we will migrate the example to use remoteproc cdev on kernel 6.6.

Regards,

Nick

Hi Nick,

   I'm sorry I forgot to introduce some background to you in advance. We are porting our dToF sensor (as a v4l2 camera sensor sub device) to TI platform, the internal R5 MCU need to control some hardware component, while my Linux side driver will control other components. My Linux side driver also need communicate with the R5 MCU to get its status or send some commands to it, we defined some business layer registers for this communications, so here "MCU registers" only means our custom registers, not real MCU registers, so here the register actually means a MCU side variable which can be read or writen by Linux side driver by RpMsg driver. I hope you can understand what I'm saying. 

    Yes, sometimes, not so often, I can see rpmsg timeout even if Linux is reading/write one register.  I will add some mutex/lock for Rpmsg in my Linux side driver to see whether it will be improved. It sounds a little related with the  concurrency.

Thanks and Best Regards

David

Nick Saulnier said:

For an example of a shared memory region, please refer to the zerocopy example:
https://git.ti.com/cgit/rpmsg/rpmsg_char_zerocopy/ 

Hi Nick,

   I noticed in this example, the Linux side code is based on Linux user space, since my Rpmsg communication is based on kernel space, would you please help to provide a example on kernel space?

Thanks and Best regards,

David

Hello David,

Your thread is assigned to someone else for now (the electricity and internet is out for my part of the city and probably will be for the next couple days - I had to drive to another town to type this response).

We don't currently provide an example of shared memory with the Linux kernel driver, though there should be other examples online (e.g., I've seen character drivers used to expose memory up to userspace - no reason they couldn't just use the memory directly from kernel space as well). The generic kernel RPMsg driver example is in the kernel under samples/rpmsg.

Regards,

Nick

Hi Nick,

    Thanks for your support and update.

Thanks and Best regards,

David

Hi David,

Is there anything pending, or can we close this thread?

Thanks

Yogesh