On 2006-08-09, Linards Ticmanis wrote:

> is there in Linux any kind of driver (or driver support infrastructure)
> for the DCE end of a RS232 - basically, for running Linux inside a
> modem-like device and driving its RS232 interface.
>
> This means especially that the DCD (Data Carrier Detect) and RI (Ring
> Indicator) lines must be *output* lines settable by software, not input
> lines. It would be best if I could simply use the normal tty_ioctl(4)
> interface for that.

I've been messing around with Linux serial stuff for a lot of
years and I've never come across a driver like that.

> I know that there's a concept of "gadget drivers" in the Linux-USB
> infrastructure, for running Linux in an USB device. Is there such a
> thing for RS232 as well? Or will I have to write my own?

You'll have to write your own anyway, since your hardware is
unique, right?

--
Grant Edwards grante Yow! I selected E5... but
at I didn't hear "Sam the Sham
visi.com and the Pharoahs"!

> This means especially that the DCD (Data Carrier Detect) and RI (Ring
> Indicator) lines must be *output* lines settable by software, not input
> lines.

Those don't need to be fast and thus can be set from user land using
some bit-I/O driver.

-Michael

Hello Grant,

Grant Edwards schrieb:

> > is there in Linux any kind of driver (or driver support infrastructure)
> > for the DCE end of a RS232 - basically, for running Linux inside a
> > modem-like device and driving its RS232 interface.
> >
> > This means especially that the DCD (Data Carrier Detect) and RI (Ring
> > Indicator) lines must be *output* lines settable by software, not input
> > lines. It would be best if I could simply use the normal tty_ioctl(4)
> > interface for that.
>
> I've been messing around with Linux serial stuff for a lot of
> years and I've never come across a driver like that.

Ok, thanks, saves me a bunch of searching.

> > I know that there's a concept of "gadget drivers" in the Linux-USB
> > infrastructure, for running Linux in an USB device. Is there such a
> > thing for RS232 as well? Or will I have to write my own?
>
> You'll have to write your own anyway, since your hardware is
> unique, right?

Not quite, it's using an Atmel AT91 CPU, which is an ARM with built in
UART (and other stuff); a "normal" (DTE-end) driver for this UART is
already in the standard kernel sources.

Thanks and best wishes,
--
Linards Ticmanis

On 2006-08-09, Linards Ticmanis wrote:
> Hello Grant,
>
> Grant Edwards schrieb:
>
>> > is there in Linux any kind of driver (or driver support infrastructure)
>> > for the DCE end of a RS232 - basically, for running Linux inside a
>> > modem-like device and driving its RS232 interface.
>> >
>> > This means especially that the DCD (Data Carrier Detect) and RI (Ring
>> > Indicator) lines must be *output* lines settable by software, not input
>> > lines. It would be best if I could simply use the normal tty_ioctl(4)
>> > interface for that.
>>
>> I've been messing around with Linux serial stuff for a lot of
>> years and I've never come across a driver like that.
>
> Ok, thanks, saves me a bunch of searching.
>
>> > I know that there's a concept of "gadget drivers" in the Linux-USB
>> > infrastructure, for running Linux in an USB device. Is there such a
>> > thing for RS232 as well? Or will I have to write my own?
>>
>> You'll have to write your own anyway, since your hardware is
>> unique, right?
>
> Not quite, it's using an Atmel AT91 CPU, which is an ARM with built in
> UART (and other stuff); a "normal" (DTE-end) driver for this UART is
> already in the standard kernel sources.

It shouldn't be difficult to modify that driver to treat RTS
and DTR as inputs and CTS, DCD, DSR, RI as outputs. I don't
think the tty layer cares -- it just passes the bitmasks back
and forth between userspace and the low-level driver.

--
Grant Edwards grante Yow! Of course, you
at UNDERSTAND about the PLAIDS
visi.com in the SPIN CYCLE --

Grant Edwards schrieb:

> > Not quite, it's using an Atmel AT91 CPU, which is an ARM with built in
> > UART (and other stuff); a "normal" (DTE-end) driver for this UART is
> > already in the standard kernel sources.
>
> It shouldn't be difficult to modify that driver to treat RTS
> and DTR as inputs and CTS, DCD, DSR, RI as outputs. I don't
> think the tty layer cares -- it just passes the bitmasks back
> and forth between userspace and the low-level driver.

Unfortunately, it does care a bit - it removes bits that belong to the
"wrong" direction in drivers/char/tty_io.c, function tty_tiocmset.
Shouldn't be too hard to patch, though.

--
Linards Ticmanis

On 2006-08-09, Linards Ticmanis wrote:
> Grant Edwards schrieb:
>
>>> Not quite, it's using an Atmel AT91 CPU, which is an ARM with
>>> built in UART (and other stuff); a "normal" (DTE-end) driver
>>> for this UART is already in the standard kernel sources.
>>
>> It shouldn't be difficult to modify that driver to treat RTS
>> and DTR as inputs and CTS, DCD, DSR, RI as outputs. I don't
>> think the tty layer cares -- it just passes the bitmasks back
>> and forth between userspace and the low-level driver.
>
> Unfortunately, it does care a bit

You're right. It oughtn't, but it does.

> - it removes bits that belong to the "wrong" direction in
> drivers/char/tty_io.c, function tty_tiocmset. Shouldn't be too
> hard to patch, though.

There's no excuse not to make the tty layer DCE/DTE agnostic.
The low-level driver's tiocmset() method should be quite
capable of ignoring bits it can't control (and all the ones
I've ever looked at certainly do).

The tty layer is throwing away perfectly good bits for no
reason.

--
Grant Edwards grante Yow! ... I have read the
at INSTRUCTIONS...
visi.com

Grant Edwards schrieb:

> There's no excuse not to make the tty layer DCE/DTE agnostic.
> The low-level driver's tiocmset() method should be quite
> capable of ignoring bits it can't control (and all the ones
> I've ever looked at certainly do).
>
> The tty layer is throwing away perfectly good bits for no
> reason.

And there's a lot more badness in drivers/serial/serial_core.c, though
this arguably is already a part of the low-level code (or mabye one
should call it mid-level).

--
Linards Ticmanis

On 9 Aug 2006 06:51:55 -0700, "Linards Ticmanis"
wrote:

>is there in Linux any kind of driver (or driver support infrastructure)
>for the DCE end of a RS232 - basically, for running Linux inside a
>modem-like device and driving its RS232 interface.
>
>This means especially that the DCD (Data Carrier Detect) and RI (Ring
>Indicator) lines must be *output* lines settable by software, not input
>lines. It would be best if I could simply use the normal tty_ioctl(4)
>interface for that.

Since you have already rewired the UART Tx to pin 2 on a 9 pin
D-connector and UART Rx to pin 3, for simplified RS-232 handshaking,
why don't you rewire e.g. the DTR UART bit to DCD D-connector pin and
CTS UART bit to the RI pin on the D-connector. The DSR connector pin
could be driven from the DCE power supply +12 V through a resistor.
The software would have to control the DTR and DCD settings to alter
the DCD and RI lines on the serial card.

However, if full DCE RS-232 handshaking is needed, at least the DTR,
RTS inputs and DSR, CTS, DCD and RI outputs would be needed and it is
unlikely that any ordinary UART would have 2 input and 4 output
programmable control lines.

In practice, you would need to assign a separate (parallel) port on
your micro controller and attach a suitable number of RS-232/TTL and
TTL/RS-232 converters to that port and program the data direction
register. Thus, this port would be separate from the serial port and
could easily be controlled by user mode software.

Paul

On 2006-08-10, Linards Ticmanis wrote:
> Grant Edwards schrieb:
>
>> There's no excuse not to make the tty layer DCE/DTE agnostic.
>> The low-level driver's tiocmset() method should be quite
>> capable of ignoring bits it can't control (and all the ones
>> I've ever looked at certainly do).
>>
>> The tty layer is throwing away perfectly good bits for no
>> reason.
>
> And there's a lot more badness in drivers/serial/serial_core.c, though
> this arguably is already a part of the low-level code (or mabye one
> should call it mid-level).

I guess I would consider it a block of comon low-level code,
and you're right: it assumes it's DTE (e.g. it tries to set set
RTS+DTR on open).

If I were to write a DCE driver, I'd probably not use
serial_core.c. FWIW, there are still serial drivers that don't
use serial_core.c, since it makes some pretty restrictive
assumptions about how the serial interface hardware works.

--
Grant Edwards grante Yow! I'm a nuclear
at submarine under the
visi.com polar ice cap and I need
a Kleenex!

>
> And there's a lot more badness in drivers/serial/serial_core.c, though
> this arguably is already a part of the low-level code (or mabye one
> should call it mid-level).
>

BTW.:

In this directory I see this file and several files for different
hardware setups.

In /drivers/char there is rtc.c. This seems to be just for the PC based
clock chip but not usable for (e.g.) ARM (as it e.g. uses "CMOS_WRITE"
and friends). How can it be "legal" that there is no hint in the
directory or file name that this is dedicated for a certain class of
hardware ?

Would it not make sense to use a similar two level method as with
serial_core.c ?

I am investigating this, as I am considering doing a "multiple-timer"
driver that uses the RTC (on PC) or some hardware timer (on ARM) as a
quite fast interrupt source and allows multiple user land processes
(threads) to wait for periodic events with an individually definable
frequency, but with certain (somewhat obscure ) mutual synchronizing
features. I'd rater use some low level module to hook the interrupt than
just copy the code from the hardware dependent driver into my module.

-Michael