[okl4-developer] serial driver l4_kdb

[Lukas HANEL]

Hi Josh,

thanks a lot for your comprehensive answer!

So in fact I had already written my own iguana level serial driver for
my device and was searching ways to test its functionality.

After your response I used the posix example and could see that output
works!

The aforementioned problem with the kernel emptying the fifo disapeared
after a complete rebuild.

Lukas Hänel

Josh Matthews wrote:
> Hi Lukas,
>
> On Sat, May 24, 2008 at 1:32 AM, Lukas HANEL <lukas.hanel at st.com
> <mailto:lukas.hanel at st.com>> wrote:
>
>     Geoffrey Lee wrote:
>     > Output via kdb can go out in one of the following two ways:
>     >
>     > (1) your printf() or whatever libc print function gets routed to
>     > libs/c/sys-*/ which ends up calling the kdb character print
>     > functions, or you implement your own such as print function
>     > and end up calling the kdb character print functions directly
>     >
>
>     In this case, how/where is the kernel/userland boundary crossed?
>
>
> This depends on the implementation that the write_fn() and read_fn()
> function pointers point to for the stdio fd's. This setup is done in
> libs/c/src/sys-{X}/sys_stdio.c (where {X} is the build environment for
> your application). For Iguana, the read/write function pointers for
> stdout are statically initialised to point to a read_init and
> write_init function, which then points them on the first read/write to
> a read/write function which uses L4_KDB_PrintChar and L4_KDB_ReadChar.
> The L4_KDB functions (architecture-specific, defined in
> arch/{arch}/libs/l4/include/kdebug.h) trap into the kernel to perform
> the read/write - it is here that the user/kernel boundary is crossed.
>  
>
>     > (2) you use vserial for your i/o.  In this instance you don't
>     > really care what the backend is.  If you choose to use l4kdb_uart,
>     > it will do a cheesy emulation of a serial device, enough to
>     > read and write, anyway.
>     >
>
>     Using vserial for my i/o means I do not use printf?
>     In fact I am now testing the functionality of my serial driver and
>     have
>     not yet found a way to send data via the vserial interface. How
>     would I
>     do this?
>
>
> No, you still use printf(). printf() eventually gets routed through to
> fputc() (passing stdout as the output stream), which calls the
> write_fn function pointer on stdout - as mentioned, that function
> pointer gets set up in libs/c/src/sys-{X}/sys_stdio.c, and you can
> point it to whatever you want. In this case, you would point it to a
> write function that would use the vserial interface to output the
> character. In fact, this has already been done -  check out
> libs/c/src/sys-posix/sys_stdio.c. There are two functions  -
> serial_read and serial_write - which would be good to look at as a
> starting point to understanding how this is done. When you build an
> example in the POSIX environment (using example_env=posix), all output
> is over vserial (unless vserial is not available, in which case it
> falls back to KDB). Let us know if you have any questions about that.
>  
>
>     Also, I am facing a problem with the kernel console. Iguana and the
>     Kernel use the same serial device, like in gumstix. When I send a
>     char,
>     I get an interrupt delivered to my driver, but the data register
>     of the
>     device is already empty. If, in the kernel's console.cc/get_c()
>     function
>     I disable the reading of this register, it works. I programmed the
>     kernel timer for a 10ms tick.
>     Do you have an idea what I could do? E.g. is it possible to have
>     kernel
>     and iguana use the same serial device?
>
>
> The problem you are seeing is a result of both the kernel and iguana
> using KDB for IO - I'd definitely recommend switching over to using a
> "real" serial device for applications.
>
> Best regards,
> Josh Matthews
>