tty_tiocmset() masks out TIOCM_RI and TIOCM_CD, which breaks rfcomm

tty_tiocmset() masks out TIOCM_RI and TIOCM_CD, which breaks rfcomm

[Adam Bliss]

Hello. This is my first email to lkml so please be gentle.

In drivers/char/tty_io.c:2477, the tty_tiocmset() function masks out
all bits except these:

   TIOCM_DTR|TIOCM_RTS|TIOCM_OUT1|TIOCM_OUT2|TIOCM_LOOP

[1] I do not understand the purpose of this masking; it seems to have
existed as long as tty_tiocmset() [2], though TIOCM_LOOP was added to
the whitelist later [3].  I wonder if it would be appropriate to add
TIOCM_CD and TIOCM_RI to the list?  Like this:

--- old/drivers/char/tty_io.c       2009-01-20 14:35:58.000000000 -0800
+++ new/drivers/char/tty_io.c       2009-01-20 14:39:43.000000000 -0800
@@ -2474,8 +2474,8 @@
                clear = ~val;
                break;
        }
-       set &= TIOCM_DTR|TIOCM_RTS|TIOCM_OUT1|TIOCM_OUT2|TIOCM_LOOP;
-       clear &= TIOCM_DTR|TIOCM_RTS|TIOCM_OUT1|TIOCM_OUT2|TIOCM_LOOP;
+       set &= TIOCM_DTR|TIOCM_RTS|TIOCM_CD|TIOCM_RI|TIOCM_OUT1|TIOCM_OUT2|TIOCM_LOOP;
+       clear &=
TIOCM_DTR|TIOCM_RTS|TIOCM_CD|TIOCM_RI|TIOCM_OUT1|TIOCM_OUT2|TIOCM_LOOP;
        return tty->ops->tiocmset(tty, file, set, clear);
 }


Traditionally, RI (Ring Indicator) and CD (Carrier Detect) were
typically used for DCE -> DTE communication, so perhaps this is why
they were left off initially.  However, when Bluetooth is used to
create a rfcomm serial connection, there is no real distinction
between DTE and DCE, and these signals can be used in either
direction.  The rfcomm driver correctly handles these signals in its
tiocmset function (see net/blutooth/rfcomm/tty.c lines 1033-1036 [4]),
but it can never receive these bits from the tty subsystem because of
the mask above.

My use case is for a linux desktop to communicate with an NXP BGB203
Bluetooth serial chip.  The chip provides outputs for the RI and CD
signals, but I cannot find any way to send these signals through the
stock rfcomm tty driver.

It might be appropriate to include other signals as well, such as CTS
and DSR.  However, these do not seem necessary for rfcomm, since the
rfcomm driver will interpret a DTR as a DSR and an RTS as a CTS.  My
main concern is for RI and CD, for which there does not seem to be a
workaround.

(Before asking this question I have consulted ESR's "How To Ask
Questions".  Google searches for TIOCM_RI have not turned up any
information on this issue, in the lkml archives or elsewhere.  The
initial commit of tiocmset does not explain the reasoning behind the
mask [2], so I can't determine if it's inappropriate to include RI and
CD.  I initially posted this question to the linux-bluetooth mailing
list [5] but Marcel Holtmann advised me to ask again here, for which I
thank him.  He suggested I include Alan Cox, and I am including
Russell King because he seems to have written the mask in the first
place [2].)

Thanks in advance,
--Adam Bliss

References:
1: <http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=drivers/char/tty_io.c;hb=fc6f6238226e6d1248e1967eae2bf556eaf3ac17>
or http://a.gd/c4c55a
2: <http://git.kernel.org/?p=linux/kernel/git/tglx/history.git;a=commitdiff;h=3bbb0896d422a2c5d5c6f8eb584e9bda3d3277a5>
or http://a.gd/283b12
3: <http://git.kernel.org/?p=linux/kernel/git/tglx/history.git;a=blobdiff;f=drivers/char/tty_io.c;h=357c331adc8d61477092f002e6e3ed23838e1d5a;hp=b82dd71948b8b3fff0f8996b78ae09fb162de82b;hb=276fccf5526400db75752ed7ce8cbf8bb68591ca;hpb=2c0bec026ccb24b6a8f31138471b79f5efcb0168>
or http://a.gd/da5064
4: <http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=net/bluetooth/rfcomm/tty.c;hb=09c7d8293a2d1317d16ef4ddb9f6dd2553d0694e>
or http://a.gd/ec3a81
5: <http://article.gmane.org/gmane.linux.bluez.kernel/1229> or
http://a.gd/2e7a0a

Re: tty_tiocmset() masks out TIOCM_RI and TIOCM_CD, which breaks rfcomm

[Alan Cox]

> TIOCM_CD and TIOCM_RI to the list?  Like this:

CD and RI are inputs which is why they are masked.

> Traditionally, RI (Ring Indicator) and CD (Carrier Detect) were
> typically used for DCE -> DTE communication, so perhaps this is why
> they were left off initially.  However, when Bluetooth is used to

The cables end up doing the conversion so the PC world always sees them
from the one end - even if they are in fact crossed over in the cable.

> It might be appropriate to include other signals as well, such as CTS
> and DSR.  However, these do not seem necessary for rfcomm, since the
> rfcomm driver will interpret a DTR as a DSR and an RTS as a CTS.  My
> main concern is for RI and CD, for which there does not seem to be a
> workaround.

I would have expected the driver to be remapping RI/CD just as it does
for the DTR<->DSR RTS<->CTS cable switch around.

It's certainly possible to fix it, but that would need a double check on
the existing drivers and that passing new suprise bits won't do any harm.
The real question is what should the right behaviour be - and to me that
seems to be to logically flip all the lines not just DTR/DSR RTS/CTS

Re: tty_tiocmset() masks out TIOCM_RI and TIOCM_CD, which breaks rfcomm

[Adam Bliss]

Hi, Alan.  Thanks for the reply.

On Wed, Jan 21, 2009 at 07:32, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
> I would have expected the driver to be remapping RI/CD just as it does
> for the DTR<->DSR RTS<->CTS cable switch around.
>
> It's certainly possible to fix it, but that would need a double check on
> the existing drivers and that passing new suprise bits won't do any harm.
> The real question is what should the right behaviour be - and to me that
> seems to be to logically flip all the lines not just DTR/DSR RTS/CTS

What would you expect the driver to map RI and CD to?  I don't think
there are DTE->DCE signals that correspond to RI and CD.  Possibly
OUT1 and OUT2 could be used, since I don't think these are used
anywhere else in rfcomm, but this might be unintuitive.  I'll let
Marcel address that idea.

--Adam