how CAN-bus Linux driver MCP251x.c works

[Dave Hylands]

Hi Cheng,

On Fri, Jan 14, 2011 at 6:23 AM, cheng chen <freakrobot at acm.org> wrote:
> I am studying CAN-bus chip Linux driver, MCP251x.c in the kernel source.
> However, I find a little confused about how this driver works, it registers
> itself as a SPI slave device using the structure
> struct spi_driver  mcp2510_can_driver = {
>     .driver = {...}
>     .probe=
>     ...
> }
> In probe function, it registers itself as a "net_device" with
> "mcp251x_netdev_ops". So I am not sure how this system works.
> I know that the userspace must see it as a network driver. But can userspace
> really see this driver? Or what it can see is the module "can.ko" -- the
> protocol stack ?
> Does the network stack talks to SPI and SPI talks to mcp2510 chip and
> transmit messages can-bus? Or something else?

I haven't looked at the code but I have looked at the chip, and I've
been sort of following along on the Socket-CAN list.
https://lists.berlios.de/mailman/listinfo/socketcan-users

It would register itself with the network stack, and the network stack
will expect to send and receive packets. The CAN driver then needs to
send these packets to the chip over a SPI bus. So it uses the SPI
infrastructure to send these packets to the actual MCP2510.

Every system has different SPI devices, so this way, the MCP2510
driver doesn't have to know how the actual SPI layer works, it just
knows it needs to use the SPI protocol to talk to the MCP2510.

>From userspace, you would use the socket API to send you data to
whatever device you're communicating with via CAN.

So the whole thing should go together something like this:

user space talks to socket API
network driver talks to CAN driver
CAN driver talks to SPI driver
SPI driver talks to SPI hardware
messages go over the SPI bus to MCP2510
MCP2510 sends messages over the CAN bus to the actual device

Dave Hylands