[ECOS] Re: Ethernet and Serial drivers for Linux target?

[Bart Veer <bartv at cygnus dot co dot uk>]

>>>>> "Grant" == Grant Edwards <grante@visi.com> writes:

    Grant> On Thu, Nov 11, 1999 at 05:10:35PM +0000, Jonathan Larmour wrote:
    >> Grant Edwards wrote:
    >> > 
    >> > I told management that it should be possible to write serial port and
    >> > Ethernet device drivers for the Linux eCos target, thus allowing
    >> > people to do eCos application development on Linux boxes.  Of course
    >> > the performance and timing won't be the same, but for basic application
    >> > functionality it should be good enough for many things.
    >> > 
    >> > Has anybody done this?  I assume that all you have to do is to map
    >> > cyg_io_read() and cyg_io_write() into read() and write() calls, and
    >> > map cyg_io_get_config() and cyg_io_set_config() into appropriate
    >> > ioctl() calls?
    >> 
    >> One other issue that no-one else has mentioned is that you
    >> can't make calls to the system read() and write(), ioctl() etc.
    >> directly. To do that would involve linking with glibc, and that
    >> would simply not work.

    Grant> I presume I could grab the source for Linux glibc read()
    Grant> write() and ioctl(), strip out the stuff I don't need, and
    Grant> impliment my own eCos-friendly versions of those calls...

It may be better to look at syscall-i386-linux-1.0.S in the synthetic
target HAL sources. Note that if you want more realistic device driver
behaviour, with interrupts etc., then you will want to experiment with
asynchronous I/O, SIGIO signals, etc. If an eCos thread performs a
blocking Linux read() syscall then you will block the entire eCos
application, which is probably not what you want. Of course this makes
the problem a bit more challenging :-)

    >> Instead all the current interfacing is done using kernel system
    >> calls directly.

    Grant> Ah. That's one question about which I had begun to wonder.
    Grant> I haven't looked at Linux system calls very closely, but
    Grant> there shouldn't be that much of a difference in abstraction
    Grant> bewteen what the kernel provides and what I need to make an
    Grant> eCos serial driver work.

    >> Of course, this doesn't prevent you having a separate native
    >> linux program acting as a server for the client requests,
    >> communicating using fd's. And that's where Bart's solution
    >> would come in, in the thread in October he mentioned.

    Grant> I suppose I could also just use the eCos 16550 driver and
    Grant> diddle the hardware directly, but I'd rather let the Linux
    Grant> drivers worrry about that so that I can utilize other types
    Grant> of serial ports.

I would certainly recommend you avoid messing about with PC hardware
directly, except as an absolute last resort. The current synthetic
target provides a safe and stable development environment, not unlike
a sandbox. Fiddling with real hardware would upset this.

Bart Veer // eCos net maintainer