How to implement mutex for buffer(stream) writing

[Rico]

Hello,

Thanks for your work on NanoPb. I really like it!

I am struggling with realizing a protokoll via SPI. Nanopb is on the SPI slave side, the master will periodically poll the slave and read "one buffer". A double buffer system reserves one to read/write from and one rx/tx set is connected o the SPI via DMA.

Before writing to the buffer a mutex needs to be taken, so that if the master does read the spi buffer, the switch over does only happen after nanopb is done writing. Which can be implemented in a custom callback as far as I can see.

Now to the point, with this mechanism incomplete pb messages could be in a buffer given to the spi, since nanopb will make multiple calls to the stream callback if I understand correctly. Does anyone see a way to solve that on the nanopb size? Probably wrapping pb_encode_ex()?  

Otherwise I will probably need to transfer a small header in each spi packet indicating the payload length in addition to the length delimited pb messages. And assemble the messages back in the host.

[Petteri Aimonen]

Hi,

> Now to the point, with this mechanism incomplete pb messages could be in a
> buffer given to the spi, since nanopb will make multiple calls to the
> stream callback if I understand correctly. Does anyone see a way to solve
> that on the nanopb size? Probably wrapping pb_encode_ex()?

Yes, if you want complete messages, the easiest is to wrap outside pb_encode()
completely.

You could have functions open_spi_stream() that locks the mutex and returns
pb_ostream_t, and corresponding close_spi_stream() that is called when done
writing.

--
Petteri

[Rico]

Thanks for the fast response, sound good to me!

BTW, since probably a lot of people are asking here about transport level protocolls for embedded protobuf usage. Do you know about any simple transport protocols for SPI or similar? Feels so wrong to roll my own.

[Petteri Aimonen]

Hi,

I've used first byte after chip-select as length and then the master
knows read that many bytes.

Usually no point to make it too complex.

--
Petteri