libcanard polling too slow

[zubax]

libcanard polling too slow

[chemicstry]

Just a quick update if anyone encounters this problem. Increasing ChibiOS system tick frequency, as suggested by Pavel, fixed the issue and I'm no longer losing frames.

[pavel-kirienko]

Normally, you should fix this at the platform level. You could add an RX FIFO queue between your CAN driver and the library.

Here you can find an STM32F105-based bootloader (CM3 72 MHz, execution from flash) with ChibiOS which works without issues at 1 Mbps: https://github.com/Zubax/zubax_gnss/blob/3ed2388a7dbe0eae226293846b327309d414521e/bootloader/src/board/can.hpp

I am not familiar with your application, but I suppose the culprit might be right here:

/**
 * Hardware FIFO is 3 frames deep; the shortest (worst case) frame duration with acceptance filters allowing only
 * extended frames is:
 *  (1+11+1+1+18+1+2+4+0+15+1+1+1+7) = 64 bits = 64 microseconds @ 1 Mbps
 *
 * Multiply by the depth of the FIFO:
 *  64 * 3 = 192 microseconds
 *
 * Therefore, we must read the driver not less frequently than every 192 microseconds, otherwise we might be
 * losing frames due to RX overrun. Therefore we enforce that the system tick interval is less than that.
 */
static_assert((1000000 / CH_CFG_ST_FREQUENCY) < 180,
            "Minimal delay must be lower in order for the libcanard STM32 driver to work properly");

[chemicstry]

Thank you for the link with an example implementation, I think you should add that to the readme.

And yes, that indeed might have been the culprit. I will report if it has fixed the issue if I find time to revert my modifications and try again.

[pavel-kirienko]

I think it's well-covered in the README for the driver already: https://github.com/UAVCAN/libcanard/tree/master/drivers/stm32

Writing anything more specific would not be sufficiently generic, so I would rather not do that.

[chemicstry]

I mean adding the https://github.com/Zubax/zubax_gnss/ link to this repository as an example implementation of a kocherga bootloader. Sorry for the confusion :)

EDIT: oh my bad, I thought it was based on this bootloader framework, but it's not. Nervermind.