Serial ports with non connected RTS/CTS
Hi all,
currently, at least for 8250 based serial ports using of_serial, there is no way of indicating that the RTS / CTS lines, although supported by the UART, are not actually connected on the board.
This means that userspace needs to know about this and not use the CRTSCTS termios flag (otherwise no data is sent).
This is currently the case for Debian stattach for example (nettools 1.60) which unconditionally uses CRTSCTS (but the busybox version of slattach has a "-F" option that can be used to disable CRTSCTS).
Would it not be better to allow this situation to be described in the device tree?
Such an option already exists for the imx UART (""fsl,uart-has-rtscts")
of_serial already has the "auto-flow-control" DT property but, as stated in the documentation:
"The driver is allowed to detect support for the capability even
without this property."
In the past there was a (since reverted) property "has-hw-flow-control" See
06aa82e "serial: uart: add hw flow control support configuration" a6eec92 " Revert "serial: uart: add hw flow control support
configuration"
However that just provided another (redundant) way of *activating* hardware flow control and did not allow it to be disactivated.
Would a new DT property "no-rtscts" to do this be acceptable?
I think it would have to be a negative property to avoid breaking old device trees.
Since there seems to have been some to and fro on this issue I thought it would be a good idea to discuss before writing the patch...
Regards,
Martin
Hi,
Quoted 32 lines · On Wed, 27 Jan 2016 11:30:32 +0100 Martin Fuzzey wrote:
Hi all,
currently, at least for 8250 based serial ports using
of_serial, there is no way of indicating that the RTS / CTS lines, although supported by the UART, are not actually connected on the board.This means that userspace needs to know about this and not use the CRTSCTS termios flag (otherwise no data is sent).
This is currently the case for Debian stattach for example (nettools 1.60) which unconditionally uses CRTSCTS (but the busybox version of slattach has a "-F" option that can be used to disable CRTSCTS).
Would it not be better to allow this situation to be described in the device tree?
Such an option already exists for the imx UART (""fsl,uart-has-rtscts")
of_serialalready has the "auto-flow-control" DT property but, as stated in the documentation:
"The driver is allowed to detect support for the capability even
without this property."In the past there was a (since reverted) property "has-hw-flow-control" See
06aa82e "serial: uart: add hw flow control support configuration" a6eec92 " Revert "serial: uart: add hw flow control support
configuration"However that just provided another (redundant) way of *activating* hardware flow control and did not allow it to be disactivated.
Quoted 8 lines
Would a new DT property "no-rtscts" to do this be acceptable?
I think it would have to be a negative property to avoid breaking old device trees.
Since there seems to have been some to and fro on this issue I thought it would be a good idea to discuss before writing the patch...
If a driver has no support for RTS/CTS (e.g. due to the missing fsl,uart-has-rtscts property for i.MX UARTs) then trying to enable HW flowcontrol via the TCETS ioctl should fail (as is the case with the i.MX serial driver).
Thus it should be impossible to enable crtscts if the serial driver has no RTC/CTS support without any need for userspace to know about the HW details.
Lothar Wa?mann
Hi,
Quoted 8 lines · Lothar Wa?mann
If a driver has no support for RTS/CTS (e.g. due to the missing fsl,uart-has-rtscts property for i.MX UARTs) then trying to enable HW flowcontrol via the TCETS ioctl should fail (as is the case with the i.MX serial driver).
Thus it should be impossible to enable crtscts if the serial driver has no RTC/CTS support without any need for userspace to know about the HW details.
It doesn't do that for me [kernel 4.4]
The stty command to activate rtscts does return an error if the hardware doesn't support it.
# stty -F /dev/ttymxc1 crtscts stty: /dev/ttymxc1: unable to perform all requested operations
But the TCETS ioctl itselfdoes not fail:
ioctl(0, SNDCTL_TMR_STOP or TCSETSW, {B115200 -opost -isig -icanon -echo ...}) = 0ioctl(0, SNDCTL_TMR_TIMEBASE or TCGETS, {B115200 -opost -isig -icanon -echo ...}) = 0ioctl(0, SNDCTL_TMR_TIMEBASE or TCGETS, {B115200 -opost -isig -icanon -echo ...}) = 0write(2, "stty: ", 6stty: ) = 6
write(2, "/dev/ttymxc1: unable to perform "..., 56/dev/ttymxc1: unable to perform all requested operations) = 56
The error message is because stty does a TCGETS afterwards and errors if the result is different to that requested.
slattach does not compare the requested and actual configuaration
The kernel code in the imx driver is:
static void
imx_set_termios(struct uart_port *port, struct ktermios *termios,
struct ktermios *old)
{
...
if (termios->c_cflag & CRTSCTS) {
if (sport->have_rtscts) {
ucr2 &= ~UCR2_IRTS;
if (port->rs485.flags & SER_RS485_ENABLED) {
/*
* RTS is mandatory for rs485 operation, so keep
* it under manual control and keep transmitter
* disabled.
*/
if (!(port->rs485.flags &
SER_RS485_RTS_AFTER_SEND))
ucr2 |= UCR2_CTS;
} else {
ucr2 |= UCR2_CTSC;
}
} else {
termios->c_cflag &= ~CRTSCTS;
}
} else if (port->rs485.flags & SER_RS485_ENABLED)So requesting RTSCTS if the hardware doesn't support it does not generate an error but the bit is reset in the termios structure.
This makes sense to me. Userspace can ask for something - the driver does it's best and reports to userspace what it actually did. It's then up to userspace to decide if that's acceptable.
And indeed "man termios" :
"Note that *tcsetattr*() returns success if /any/ of the requested changes could be successfully carried out. Therefore, when making multiple changes it may be necessary to follow this call with a further call to *tcgetattr*() to check that all changes have been performed successfully. "
Martin
Quoted 20 lines · Martin Fuzzey
Hi all,
currently, at least for 8250 based serial ports using
of_serial, there is no way of indicating that the RTS / CTS lines, although supported by the UART, are not actually connected on the board.This means that userspace needs to know about this and not use the CRTSCTS termios flag (otherwise no data is sent).
This is currently the case for Debian stattach for example (nettools 1.60) which unconditionally uses CRTSCTS (but the busybox version of slattach has a "-F" option that can be used to disable CRTSCTS).
Would it not be better to allow this situation to be described in the device tree?
Such an option already exists for the imx UART (""fsl,uart-has-rtscts")
of_serialalready has the "auto-flow-control" DT property but, as stated in the documentation:
"The driver is allowed to detect support for the capability even without this property."In the past there was a (since reverted) property "has-hw-flow-control" See
06aa82e "serial: uart: add hw flow control support configuration" a6eec92 " Revert "serial: uart: add hw flow control support configuration"
Quoted 7 lines
However that just provided another (redundant) way of *activating* hardware flow control and did not allow it to be disactivated.
Would a new DT property "no-rtscts" to do this be acceptable?
I think it would have to be a negative property to avoid breaking old device trees.
Since there seems to have been some to and fro on this issue I thought it would be a good idea to discuss before writing the patch...
If I understand you correctly here
1) you want a way to describe the lack of CTS and RTS pins, and
2) you want the 8250 driver to mask off CRTSCTS when #1 is true.
Sounds ok to me, but I'm not sure how userspace will respond to CRTSCTS not sticking. That will need some testing.
Regards,
Peter Hurley
Quoted 16 lines · Peter Hurley
Quoted 15 lines · Martin Fuzzey
Hi all,
currently, at least for 8250 based serial ports using
of_serial, there is no way of indicating that the RTS / CTS lines, although supported by the UART, are not actually connected on the board.This means that userspace needs to know about this and not use the CRTSCTS termios flag (otherwise no data is sent).
This is currently the case for Debian stattach for example (nettools 1.60) which unconditionally uses CRTSCTS (but the busybox version of slattach has a "-F" option that can be used to disable CRTSCTS).
Would it not be better to allow this situation to be described in the device tree?
Such an option already exists for the imx UART (""fsl,uart-has-rtscts")
of_serialalready has the "auto-flow-control" DT property but, as stated in the documentation:
"The driver is allowed to detect support for the capability even without this property."
Quoted 8 lines
Quoted 8 lines
In the past there was a (since reverted) property "has-hw-flow-control" See
06aa82e "serial: uart: add hw flow control support configuration" a6eec92 " Revert "serial: uart: add hw flow control support configuration"However that just provided another (redundant) way of *activating* hardware flow control and did not allow it to be disactivated.
Would a new DT property "no-rtscts" to do this be acceptable?
Yes.
It's really a "the h/w designers left CTS tied/floating inactive by mistake" flag. What is done with RTS doesn't really matter. So just "no-cts" perhaps.
Quoted 1 line
Quoted 1 line
I think it would have to be a negative property to avoid breaking old device trees.
Agreed.
Rob