Hi Peter, NuttX has an SocketCAN interface as well. But currently only the Kinetis, i.MXRT and S32K1 MCU use this interface for their CAN driver.
There's a porting guide to interface with SocketCAN driver as well https://nuttx.apache.org/docs/latest/components/drivers/special/socketcan.html https://www.youtube.com/watch?v=1ZwL7OU30xw I would advise to use the arch/arm/src/s32k1xx/s32k1xx_flexcan.c driver as a reference. Regarding performance SocketCAN in NuttX is quite comparable to the character interface. But as Matt indicate the mailbox filtering is done on software level. You could expose the hardware filters using a ioctl just like the character driver though. I'm not sure which STM32 part you're using but NuttX already has some CAN drivers in place (albeit the NuttX character driver). It should be fairly simple to modify these driver and make it compile time configurable to use the character device interface or SocketCAN interface. Kind Regards, Peter. -----Original Message----- From: Matthew Trescott <matthewtresc...@gmail.com> Sent: Monday, 27 December 2021 20:35 To: peter.m...@mnet-mail.de Cc: dev@nuttx.apache.org Subject: [EXT] Re: SocketCAN for STM32 ? Caution: EXT Email Hi Peter, I haven't worked with NuttX's SocketCAN implementation or STM32, but I did write a "classic" char-device CAN driver for the Tiva MCUs. The reason I wrote a char-device driver is because the char-device CAN subsystem has ioctls for things like hardware message filtering and the interface supports (IIRC) hardware transmit FIFOs if desired. It's possible those features aren't relevant with 100+ MHz MCUs, but I imagine there are use-cases where it is. I also found that on the Tiva hardware it is difficult to make use of the hardware FIFO capability because you still need to get an IRQ for each message to handle bus errors, lost arbitration, etc that an application might be interested in. It's not that the SocketCAN implementation _couldn't_ support those hardware features. It just doesn't yet. Without looking at the Zephyr code, my guess is that it's probably not complicated enough to be worth trying to outright port. It would probably be just as easy if not easier to write an STM32 SocketCAN driver for NuttX and just use the Zephyr one as a reference. Just my (somewhat uninformed) 2ยข though. Kind regards, Matt On Sun, Dec 26, 2021 at 12:39 PM Peter Marx <peter.m...@mnet-mail.de> wrote: > > Hi, > I have seen Peter Van Der Per's great port of SocketCAN to NXP MCUs. On > Zephyr there is a SocketCAN for STM32 implementation. I am not deep enough > in C business to judge how difficult it would be to port that to Nuttx. > Is some insider here around who could comment or maybe works on it already? > > BR, > Peter
