Skip to contentSkip to conversation

tty/serial: Add a serial port simulator

PATCH In review [email protected] 6 participants 9 messages Mar 5, 2019 – Mar 29, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 tty/serial: Add a serial port simulator

From: Corey Minyard <[email protected]>

This creates simulated serial ports, both as echo devices and pipe devices. The driver reasonably approximates the serial port speed and simulates some modem control lines. It allows error injection and direct control of modem control lines.

---
Changes since v1:

* Fixed a build error on sparc due to a missing include.

* Fixed a build error on powerpc due to the termios mess.

* Fixed a use of an uninitialized lock at startup. I didn't realize exactly how the register worked and a callback happened that I wasn't expecting.

 Documentation/serial/serialsim.rst |  149 ++++
 MAINTAINERS                        |    7 +
 drivers/tty/serial/Kconfig         |   10 +
 drivers/tty/serial/Makefile        |    1 +
 drivers/tty/serial/serialsim.c     | 1101 ++++++++++++++++++++++++++++
 include/uapi/linux/serialsim.h     |   32 +
 6 files changed, 1300 insertions(+)
 create mode 100644 Documentation/serial/serialsim.rst
 create mode 100644 drivers/tty/serial/serialsim.c
 create mode 100644 include/uapi/linux/serialsim.h
Documentation/serial/serialsim.rst new file +149 -0
new file mode 100644
index 000000000000..655e10b4908e
@@ -0,0 +1,149 @@
+.. SPDX-License-Identifier: GPL-2.0+
+=====================================
+serialsim - A kernel serial simualtor
+=====================================
+
+:Author: Corey Minyard <[email protected]> / <[email protected]>
+
+The serialsim device is a serial simulator with echo and pipe devices.
+It is quite useful for testing programs that use serial ports.
+
+This attempts to emulate a basic serial device.  It uses the baud rate
+and sends the bytes through the loopback or pipe at approximately the
+speed it would on a normal serial device.
+
+There is a python interface to the special ioctls for controlling the
+remote end of the termios in addition to the standard ioctl interface
+documented below.  See https://github.com/cminyard/serialsim
+
+=====
+Using
+=====
+
+The serialsim.ko module creates two types of devices.  Echo devices
+simply echo back the data to the same device.  These devices will
+appear as /dev/ttyEcho<n>.
+
+Pipe devices will transfer the data between two devices.  The
+devices will appear as /dev/ttyPipeA<n> and /dev/ttyPipeB<n>.  And
+data written to PipeA reads from PipeB, and vice-versa.
+
+You may create an arbitrary number of devices by setting the
+nr_echo ports and nr_pipe_ports module parameters.  The default is
+four for both.
+
+This driver supports modifying the modem control lines and
+injecting various serial errors.  It also supports a simulated null
+modem between the two pipes, or in a loopback on the echo device.
+
+By default a pipe or echo comes up in null modem configuration,
+meaning that the DTR line is hooked to the DSR and CD lines on the
+other side and the RTS line on one side is hooked to the CTS line
+on the other side.
+
+The RTS and CTS lines don't currently do anything for flow control.
+
+You can modify null modem and control the lines individually
+through an interface in /sys/class/tty/ttyECHO<n>/ctrl,
+/sys/class/tty/ttyPipeA<n>/ctrl, and
+/sys/class/tty/ttyPipeB<n>/ctrl.  The following may be written to
+those files:
+
+[+-]nullmodem
+    enable/disable null modem
+
+[+-]cd
+    enable/disable Carrier Detect (no effect if +nullmodem)
+
+[+-]dsr
+    enable/disable Data Set Ready (no effect if +nullmodem)
+
+[+-]cts
+    enable/disable Clear To Send (no effect if +nullmodem)
+
+[+-]ring
+    enable/disable Ring
+
+frame
+    inject a frame error on the next byte
+
+parity
+    inject a parity error on the next byte
+
+overrun
+    inject an overrun error on the next byte
+
+The contents of the above files has the following format:
+
+tty[Echo|PipeA|PipeB]<n>
+    <mctrl values>
+
+where <mctrl values> is the modem control values above (not frame,
+parity, or overrun) with the following added:
+
+[+-]dtr
+    value of the Data Terminal Ready
+
+[+-]rts
+    value of the Request To Send
+
+The above values are not settable through this interface, they are
+set through the serial port interface itself.
+
+So, for instance, ttyEcho0 comes up in the following state::
+
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem -cd -dsr -cts -ring -dtr -rts
+
+If something connects, it will become::
+
+   ttyEcho0: +nullmodem +cd +dsr +cts -ring +dtr +rts
+
+To enable ring::
+
+   # echo "+ring" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem +cd +dsr +cts +ring +dtr +rts
+
+Now disable NULL modem and the CD line::
+
+   # echo "-nullmodem -cd" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: -nullmodem -cd -dsr -cts +ring -dtr -rts
+
+Note that these settings are for the side you are modifying.  So if
+you set nullmodem on ttyPipeA0, that controls whether the DTR/RTS
+lines from ttyPipeB0 affect ttyPipeA0.  It doesn't affect ttyPipeB's
+modem control lines.
+
+The PIPEA and PIPEB devices also have the ability to set these
+values for the other end via an ioctl.  The following ioctls are
+available:
+
+TIOCSERSNULLMODEM
+    Set the null modem value, the arg is a boolean.
+
+TIOCSERSREMMCTRL
+    Set the modem control lines, bits 16-31 of the arg is
+    a 16-bit mask telling which values to set, bits 0-15 are the
+    actual values.  Settable values are TIOCM_CAR, TIOCM_CTS,
+    TIOCM_DSR, and TIOC_RNG.  If NULLMODEM is set to true, then only
+    TIOC_RNG is settable.  The DTR and RTS lines are not here, you can
+    set them through the normal interface.
+
+TIOCSERSREMERR
+    Send an error or errors on the next sent byte.  arg is
+    a bitwise OR of (1 << TTY_xxx).  Allowed errors are TTY_BREAK,
+    TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.
+
+TIOCSERGREMTERMIOS
+    Return the termios structure for the other side of the pipe.
+    arg is a pointer to a standard termios struct.
+
+TIOCSERGREMRS485
+    Return the remote RS485 settings, arg is a pointer to a struct
+    serial_rs485.
+
+Note that unlike the sysfs interface, these ioctls affect the other
+end.  So setting nullmodem on the ttyPipeB0 interface sets whether
+the DTR/RTS lines on ttyPipeB0 affect ttyPipeA0.
MAINTAINERS +7 -0
index 9919840d54cd..aa1eb3c07f44 100644
@@ -13702,6 +13702,13 @@ L:	[email protected]
 S:	Maintained
 F:	drivers/media/rc/serial_ir.c
 
+SERIAL PORT SIMULATOR
+M:	Corey Minyard <[email protected]>
+S:	Maintained
+F:	Documentation/serial/serialsim.rst
+F:	drivers/tty/serial/serialsim.c
+F:	include/uapi/linux/serialsim.h
+
 SFC NETWORK DRIVER
 M:	Solarflare linux maintainers <[email protected]>
 M:	Edward Cree <[email protected]>
drivers/tty/serial/Kconfig +10 -0
index 089a6f285d5e..789113055e2f 100644
@@ -1565,4 +1565,14 @@ endmenu
 config SERIAL_MCTRL_GPIO
 	tristate
 
+config SERIAL_SIMULATOR
+	tristate "Serial port simulator"
+	default n
+	help
+	  Support for a simulated device that behaves like a normal
+	  serial port.  It has echo devices and pipe devices.  This
+	  is useful for testing programs that use serial ports.
+
+	  If unsure, say N.
+
 endif # TTY
drivers/tty/serial/Makefile +1 -0
index 1511e8a9f856..dbd9ee0f1c6b 100644
@@ -91,6 +91,7 @@ obj-$(CONFIG_SERIAL_PIC32)	+= pic32_uart.o
 obj-$(CONFIG_SERIAL_MPS2_UART)	+= mps2-uart.o
 obj-$(CONFIG_SERIAL_OWL)	+= owl-uart.o
 obj-$(CONFIG_SERIAL_RDA)	+= rda-uart.o
+obj-$(CONFIG_SERIAL_SIMULATOR)	+= serialsim.o
 
 # GPIOLIB helpers for modem control lines
 obj-$(CONFIG_SERIAL_MCTRL_GPIO)	+= serial_mctrl_gpio.o
drivers/tty/serial/serialsim.c new file +1101 -0

1,101 lines not on this page · raw message

include/uapi/linux/serialsim.h new file +32 -0
new file mode 100644
index 000000000000..34369f1a85b0
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+
+/*
+ * serialsim - Emulate a serial device in a loopback and/or pipe
+ */
+
+/*
+ * TTY IOCTLs for controlling the modem control and for error injection.
+ * See drivers/tty/serial/serialsim.c and Documentation/serial/serialsim.rst
+ * for details.
+ */
+
+#ifndef LINUX_SERIALSIM_H
+#define LINUX_SERIALSIM_H
+
+#include <linux/ioctl.h>
+#include <asm/termbits.h>
+
+#define TIOCSERSREMNULLMODEM	0x54e4
+#define TIOCSERSREMMCTRL	0x54e5
+#define TIOCSERSREMERR		0x54e6
+#ifdef TCGETS2
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios2)
+#else
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios)
+#endif
+#define TIOCSERGREMNULLMODEM	_IOR('T', 0xe8, int)
+#define TIOCSERGREMMCTRL	_IOR('T', 0xe9, unsigned int)
+#define TIOCSERGREMERR		_IOR('T', 0xea, unsigned int)
+#define TIOCSERGREMRS485	_IOR('T', 0xeb, struct serial_rs485)
+
+#endif
Signature · 1 line
2.17.1
reviewer
#1 lore raw

Hi Corey,

Just some doc comments.

Documentation/serial/serialsim.rst line 3

Quoted 9 lines · On 3/5/19 9:12 AM, [email protected] wrote:
Documentation/serial/serialsim.rst new file +3 -0
new file mode 100644
index 000000000000..655e10b4908e
@@ -0,0 +1,149 @@
+.. SPDX-License-Identifier: GPL-2.0+
+=====================================
+serialsim - A kernel serial simualtor

xxxxxxxxx
serial device simulator

Quoted 15 lines
Documentation/serial/serialsim.rst new file +15 -0
+=====================================
+
+:Author: Corey Minyard <[email protected]> / <[email protected]>
+
+The serialsim device is a serial simulator with echo and pipe devices.
+It is quite useful for testing programs that use serial ports.
+
+This attempts to emulate a basic serial device.  It uses the baud rate
+and sends the bytes through the loopback or pipe at approximately the
+speed it would on a normal serial device.
+
+There is a python interface to the special ioctls for controlling the
+remote end of the termios in addition to the standard ioctl interface
+documented below.  See https://github.com/cminyard/serialsim
+

Documentation/serial/serialsim.rst line 28

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+=====
+Using
+=====
+
+The serialsim.ko module creates two types of devices.  Echo devices
+simply echo back the data to the same device.  These devices will
+appear as /dev/ttyEcho<n>.
+
+Pipe devices will transfer the data between two devices.  The
+devices will appear as /dev/ttyPipeA<n> and /dev/ttyPipeB<n>.  And

Any

Documentation/serial/serialsim.rst line 32

Quoted 4 lines
Documentation/serial/serialsim.rst new file +4 -0
+data written to PipeA reads from PipeB, and vice-versa.
+
+You may create an arbitrary number of devices by setting the
+nr_echo ports and nr_pipe_ports module parameters.  The default is

nr_echo_ports

Documentation/serial/serialsim.rst line 33

Quoted 1 line
Documentation/serial/serialsim.rst new file +1 -0
+four for both.

or for each.

Quoted 33 lines
Documentation/serial/serialsim.rst new file +33 -0
+
+This driver supports modifying the modem control lines and
+injecting various serial errors.  It also supports a simulated null
+modem between the two pipes, or in a loopback on the echo device.
+
+By default a pipe or echo comes up in null modem configuration,
+meaning that the DTR line is hooked to the DSR and CD lines on the
+other side and the RTS line on one side is hooked to the CTS line
+on the other side.
+
+The RTS and CTS lines don't currently do anything for flow control.
+
+You can modify null modem and control the lines individually
+through an interface in /sys/class/tty/ttyECHO<n>/ctrl,
+/sys/class/tty/ttyPipeA<n>/ctrl, and
+/sys/class/tty/ttyPipeB<n>/ctrl.  The following may be written to
+those files:
+
+[+-]nullmodem
+    enable/disable null modem
+
+[+-]cd
+    enable/disable Carrier Detect (no effect if +nullmodem)
+
+[+-]dsr
+    enable/disable Data Set Ready (no effect if +nullmodem)
+
+[+-]cts
+    enable/disable Clear To Send (no effect if +nullmodem)
+
+[+-]ring
+    enable/disable Ring
+

Documentation/serial/serialsim.rst line 76

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+frame
+    inject a frame error on the next byte
+
+parity
+    inject a parity error on the next byte
+
+overrun
+    inject an overrun error on the next byte
+
+The contents of the above files has the following format:

have

Quoted 41 lines
Documentation/serial/serialsim.rst new file +41 -0
+
+tty[Echo|PipeA|PipeB]<n>
+    <mctrl values>
+
+where <mctrl values> is the modem control values above (not frame,
+parity, or overrun) with the following added:
+
+[+-]dtr
+    value of the Data Terminal Ready
+
+[+-]rts
+    value of the Request To Send
+
+The above values are not settable through this interface, they are
+set through the serial port interface itself.
+
+So, for instance, ttyEcho0 comes up in the following state::
+
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem -cd -dsr -cts -ring -dtr -rts
+
+If something connects, it will become::
+
+   ttyEcho0: +nullmodem +cd +dsr +cts -ring +dtr +rts
+
+To enable ring::
+
+   # echo "+ring" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem +cd +dsr +cts +ring +dtr +rts
+
+Now disable NULL modem and the CD line::
+
+   # echo "-nullmodem -cd" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: -nullmodem -cd -dsr -cts +ring -dtr -rts
+
+Note that these settings are for the side you are modifying.  So if
+you set nullmodem on ttyPipeA0, that controls whether the DTR/RTS
+lines from ttyPipeB0 affect ttyPipeA0.  It doesn't affect ttyPipeB's
+modem control lines.

Documentation/serial/serialsim.rst line 127

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+
+The PIPEA and PIPEB devices also have the ability to set these
+values for the other end via an ioctl.  The following ioctls are
+available:
+
+TIOCSERSNULLMODEM
+    Set the null modem value, the arg is a boolean.
+
+TIOCSERSREMMCTRL
+    Set the modem control lines, bits 16-31 of the arg is

are

Documentation/serial/serialsim.rst line 136

Quoted 9 lines
Documentation/serial/serialsim.rst new file +9 -0
+    a 16-bit mask telling which values to set, bits 0-15 are the
+    actual values.  Settable values are TIOCM_CAR, TIOCM_CTS,
+    TIOCM_DSR, and TIOC_RNG.  If NULLMODEM is set to true, then only
+    TIOC_RNG is settable.  The DTR and RTS lines are not here, you can
+    set them through the normal interface.
+
+TIOCSERSREMERR
+    Send an error or errors on the next sent byte.  arg is
+    a bitwise OR of (1 << TTY_xxx).  Allowed errors are TTY_BREAK,

is this better: (or I don't understand?)
a bitwise OR of (1 << TTY_xxx) and one (or more of) the allowed error flags TTY_BREAK, TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.

Documentation/serial/serialsim.rst line 149

Quoted 13 lines
Documentation/serial/serialsim.rst new file +13 -0
+    TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.
+
+TIOCSERGREMTERMIOS
+    Return the termios structure for the other side of the pipe.
+    arg is a pointer to a standard termios struct.
+
+TIOCSERGREMRS485
+    Return the remote RS485 settings, arg is a pointer to a struct
+    serial_rs485.
+
+Note that unlike the sysfs interface, these ioctls affect the other
+end.  So setting nullmodem on the ttyPipeB0 interface sets whether
+the DTR/RTS lines on ttyPipeB0 affect ttyPipeA0.

cheers.

Signature · 1 line
~Randy
#2 lore raw
Quoted 3 lines · Randy Dunlap

Hi Corey,

Just some doc comments.

Thanks a bunch. A few comments inline on things I didn't do quite like you suggested..

Quoted 90 lines · On 3/5/19 9:12 AM, [email protected] wrote:

Documentation/serial/serialsim.rst line 3

Quoted 9 lines
Documentation/serial/serialsim.rst new file +3 -0
new file mode 100644
index 000000000000..655e10b4908e
@@ -0,0 +1,149 @@
+.. SPDX-License-Identifier: GPL-2.0+
+=====================================
+serialsim - A kernel serial simualtor

xxxxxxxxx
serial device simulator

Quoted 15 lines
Documentation/serial/serialsim.rst new file +15 -0
+=====================================
+
+:Author: Corey Minyard <[email protected]> / <[email protected]>
+
+The serialsim device is a serial simulator with echo and pipe devices.
+It is quite useful for testing programs that use serial ports.
+
+This attempts to emulate a basic serial device.  It uses the baud rate
+and sends the bytes through the loopback or pipe at approximately the
+speed it would on a normal serial device.
+
+There is a python interface to the special ioctls for controlling the
+remote end of the termios in addition to the standard ioctl interface
+documented below.  See https://github.com/cminyard/serialsim
+

Documentation/serial/serialsim.rst line 28

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+=====
+Using
+=====
+
+The serialsim.ko module creates two types of devices.  Echo devices
+simply echo back the data to the same device.  These devices will
+appear as /dev/ttyEcho<n>.
+
+Pipe devices will transfer the data between two devices.  The
+devices will appear as /dev/ttyPipeA<n> and /dev/ttyPipeB<n>.  And

Any

Documentation/serial/serialsim.rst line 32

Quoted 4 lines
Documentation/serial/serialsim.rst new file +4 -0
+data written to PipeA reads from PipeB, and vice-versa.
+
+You may create an arbitrary number of devices by setting the
+nr_echo ports and nr_pipe_ports module parameters.  The default is

nr_echo_ports

Documentation/serial/serialsim.rst line 33

Quoted 1 line
Documentation/serial/serialsim.rst new file +1 -0
+four for both.

or for each.

Quoted 37 lines
Documentation/serial/serialsim.rst new file +37 -0
+
+This driver supports modifying the modem control lines and
+injecting various serial errors.  It also supports a simulated null
+modem between the two pipes, or in a loopback on the echo device.
+
+By default a pipe or echo comes up in null modem configuration,
+meaning that the DTR line is hooked to the DSR and CD lines on the
+other side and the RTS line on one side is hooked to the CTS line
+on the other side.
+
+The RTS and CTS lines don't currently do anything for flow control.
+
+You can modify null modem and control the lines individually
+through an interface in /sys/class/tty/ttyECHO<n>/ctrl,
+/sys/class/tty/ttyPipeA<n>/ctrl, and
+/sys/class/tty/ttyPipeB<n>/ctrl.  The following may be written to
+those files:
+
+[+-]nullmodem
+    enable/disable null modem
+
+[+-]cd
+    enable/disable Carrier Detect (no effect if +nullmodem)
+
+[+-]dsr
+    enable/disable Data Set Ready (no effect if +nullmodem)
+
+[+-]cts
+    enable/disable Clear To Send (no effect if +nullmodem)
+
+[+-]ring
+    enable/disable Ring
+
+frame
+    inject a frame error on the next byte
+
+parity
Quoted 8 lines

Documentation/serial/serialsim.rst line 76

Quoted 6 lines
Documentation/serial/serialsim.rst new file +6 -0
+    inject a parity error on the next byte
+
+overrun
+    inject an overrun error on the next byte
+
+The contents of the above files has the following format:

have

This intrigued me a bit. I assumed, even though "contents" can be plural, it is used in a singular fashion here because it is one "thing". So I did some research. I couldn't really find anything definitive, and there seems to be a lot of debate on this. But if you look at:
https://dictionary.cambridge.org/grammar/british-grammar/content-or-contents
you will see, when they use "contents", they use a singular verb with it:
The contents of a book is the list of chapters or articles...
So if it's good enough for Cambridge, it's good enough for me :). Though I'm certainly no grammar expert.

Quoted 46 lines
Quoted 46 lines
Documentation/serial/serialsim.rst new file +45 -0
+
+tty[Echo|PipeA|PipeB]<n>
+    <mctrl values>
+
+where <mctrl values> is the modem control values above (not frame,
+parity, or overrun) with the following added:
+
+[+-]dtr
+    value of the Data Terminal Ready
+
+[+-]rts
+    value of the Request To Send
+
+The above values are not settable through this interface, they are
+set through the serial port interface itself.
+
+So, for instance, ttyEcho0 comes up in the following state::
+
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem -cd -dsr -cts -ring -dtr -rts
+
+If something connects, it will become::
+
+   ttyEcho0: +nullmodem +cd +dsr +cts -ring +dtr +rts
+
+To enable ring::
+
+   # echo "+ring" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem +cd +dsr +cts +ring +dtr +rts
+
+Now disable NULL modem and the CD line::
+
+   # echo "-nullmodem -cd" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: -nullmodem -cd -dsr -cts +ring -dtr -rts
+
+Note that these settings are for the side you are modifying.  So if
+you set nullmodem on ttyPipeA0, that controls whether the DTR/RTS
+lines from ttyPipeB0 affect ttyPipeA0.  It doesn't affect ttyPipeB's
+modem control lines.
+
+The PIPEA and PIPEB devices also have the ability to set these
+values for the other end via an ioctl.  The following ioctls are
+available:
Quoted 8 lines

Documentation/serial/serialsim.rst line 127

Quoted 6 lines
Documentation/serial/serialsim.rst new file +6 -0
+
+TIOCSERSNULLMODEM
+    Set the null modem value, the arg is a boolean.
+
+TIOCSERSREMMCTRL
+    Set the modem control lines, bits 16-31 of the arg is

are

Same comment as above. IMHO, it's one set of bits.

Quoted 14 lines

Documentation/serial/serialsim.rst line 136

Quoted 10 lines
Documentation/serial/serialsim.rst new file +9 -0
+    a 16-bit mask telling which values to set, bits 0-15 are the
+    actual values.  Settable values are TIOCM_CAR, TIOCM_CTS,
+    TIOCM_DSR, and TIOC_RNG.  If NULLMODEM is set to true, then only
+    TIOC_RNG is settable.  The DTR and RTS lines are not here, you can
+    set them through the normal interface.
+
+TIOCSERSREMERR
+    Send an error or errors on the next sent byte.  arg is
+    a bitwise OR of (1 << TTY_xxx).  Allowed errors are TTY_BREAK,

is this better: (or I don't understand?)
a bitwise OR of (1 << TTY_xxx) and one (or more of) the allowed error flags TTY_BREAK, TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.

Well, not really. But what I wrote isn't great, so, how about:

Send an error or errors on the next sent byte. arg is a bitwise OR of (1 << TTY_BREAK), (1 << TTY_FRAME), (1 << TTY_PARITY), and (1 << TTY_OVERRUN). If none of those are set, then no error is sent.

Thanks,

-corey

Quoted 19 lines
Quoted 4 lines
Documentation/serial/serialsim.rst new file +3 -0
+    TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.
+
+TIOCSERGREMTERMIOS

Documentation/serial/serialsim.rst line 149

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+    Return the termios structure for the other side of the pipe.
+    arg is a pointer to a standard termios struct.
+
+TIOCSERGREMRS485
+    Return the remote RS485 settings, arg is a pointer to a struct
+    serial_rs485.
+
+Note that unlike the sysfs interface, these ioctls affect the other
+end.  So setting nullmodem on the ttyPipeB0 interface sets whether
+the DTR/RTS lines on ttyPipeB0 affect ttyPipeA0.

cheers.

Signature · 1 line
~Randy
reviewer
#3 lore raw
Quoted 107 lines · Corey Minyard
Quoted 3 lines · Randy Dunlap

Hi Corey,

Just some doc comments.

Thanks a bunch. A few comments inline on things I didn't do quite like you suggested..

Quoted 90 lines · On 3/5/19 9:12 AM, [email protected] wrote:

Documentation/serial/serialsim.rst line 3

Quoted 9 lines
Documentation/serial/serialsim.rst new file +3 -0
new file mode 100644
index 000000000000..655e10b4908e
@@ -0,0 +1,149 @@
+.. SPDX-License-Identifier: GPL-2.0+
+=====================================
+serialsim - A kernel serial simualtor

xxxxxxxxx
serial device simulator

Quoted 15 lines
Documentation/serial/serialsim.rst new file +15 -0
+=====================================
+
+:Author: Corey Minyard <[email protected]> / <[email protected]>
+
+The serialsim device is a serial simulator with echo and pipe devices.
+It is quite useful for testing programs that use serial ports.
+
+This attempts to emulate a basic serial device.  It uses the baud rate
+and sends the bytes through the loopback or pipe at approximately the
+speed it would on a normal serial device.
+
+There is a python interface to the special ioctls for controlling the
+remote end of the termios in addition to the standard ioctl interface
+documented below.  See https://github.com/cminyard/serialsim
+

Documentation/serial/serialsim.rst line 28

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+=====
+Using
+=====
+
+The serialsim.ko module creates two types of devices.  Echo devices
+simply echo back the data to the same device.  These devices will
+appear as /dev/ttyEcho<n>.
+
+Pipe devices will transfer the data between two devices.  The
+devices will appear as /dev/ttyPipeA<n> and /dev/ttyPipeB<n>.  And

Any

Documentation/serial/serialsim.rst line 32

Quoted 4 lines
Documentation/serial/serialsim.rst new file +4 -0
+data written to PipeA reads from PipeB, and vice-versa.
+
+You may create an arbitrary number of devices by setting the
+nr_echo ports and nr_pipe_ports module parameters.  The default is

nr_echo_ports

Documentation/serial/serialsim.rst line 33

Quoted 1 line
Documentation/serial/serialsim.rst new file +1 -0
+four for both.

or for each.

Quoted 37 lines
Documentation/serial/serialsim.rst new file +37 -0
+
+This driver supports modifying the modem control lines and
+injecting various serial errors.  It also supports a simulated null
+modem between the two pipes, or in a loopback on the echo device.
+
+By default a pipe or echo comes up in null modem configuration,
+meaning that the DTR line is hooked to the DSR and CD lines on the
+other side and the RTS line on one side is hooked to the CTS line
+on the other side.
+
+The RTS and CTS lines don't currently do anything for flow control.
+
+You can modify null modem and control the lines individually
+through an interface in /sys/class/tty/ttyECHO<n>/ctrl,
+/sys/class/tty/ttyPipeA<n>/ctrl, and
+/sys/class/tty/ttyPipeB<n>/ctrl.  The following may be written to
+those files:
+
+[+-]nullmodem
+    enable/disable null modem
+
+[+-]cd
+    enable/disable Carrier Detect (no effect if +nullmodem)
+
+[+-]dsr
+    enable/disable Data Set Ready (no effect if +nullmodem)
+
+[+-]cts
+    enable/disable Clear To Send (no effect if +nullmodem)
+
+[+-]ring
+    enable/disable Ring
+
+frame
+    inject a frame error on the next byte
+
+parity
Quoted 8 lines

Documentation/serial/serialsim.rst line 76

Quoted 6 lines
Documentation/serial/serialsim.rst new file +6 -0
+    inject a parity error on the next byte
+
+overrun
+    inject an overrun error on the next byte
+
+The contents of the above files has the following format:

have

Quoted 11 lines

This intrigued me a bit. I assumed, even though "contents" can be plural, it is used in a singular fashion here because it is one "thing". So I did some research. I couldn't really find anything definitive, and there seems to be a lot of debate on this. But if you look at:
https://dictionary.cambridge.org/grammar/british-grammar/content-or-contents
you will see, when they use "contents", they use a singular verb with it:
The contents of a book is the list of chapters or articles...
So if it's good enough for Cambridge, it's good enough for me :). Though I'm certainly no grammar expert.

OK :)

Quoted 72 lines
Quoted 46 lines
Quoted 46 lines
Documentation/serial/serialsim.rst new file +45 -0
+
+tty[Echo|PipeA|PipeB]<n>
+    <mctrl values>
+
+where <mctrl values> is the modem control values above (not frame,
+parity, or overrun) with the following added:
+
+[+-]dtr
+    value of the Data Terminal Ready
+
+[+-]rts
+    value of the Request To Send
+
+The above values are not settable through this interface, they are
+set through the serial port interface itself.
+
+So, for instance, ttyEcho0 comes up in the following state::
+
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem -cd -dsr -cts -ring -dtr -rts
+
+If something connects, it will become::
+
+   ttyEcho0: +nullmodem +cd +dsr +cts -ring +dtr +rts
+
+To enable ring::
+
+   # echo "+ring" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem +cd +dsr +cts +ring +dtr +rts
+
+Now disable NULL modem and the CD line::
+
+   # echo "-nullmodem -cd" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: -nullmodem -cd -dsr -cts +ring -dtr -rts
+
+Note that these settings are for the side you are modifying.  So if
+you set nullmodem on ttyPipeA0, that controls whether the DTR/RTS
+lines from ttyPipeB0 affect ttyPipeA0.  It doesn't affect ttyPipeB's
+modem control lines.
+
+The PIPEA and PIPEB devices also have the ability to set these
+values for the other end via an ioctl.  The following ioctls are
+available:
Quoted 8 lines

Documentation/serial/serialsim.rst line 127

Quoted 6 lines
Documentation/serial/serialsim.rst new file +6 -0
+
+TIOCSERSNULLMODEM
+    Set the null modem value, the arg is a boolean.
+
+TIOCSERSREMMCTRL
+    Set the modem control lines, bits 16-31 of the arg is

are

Same comment as above. IMHO, it's one set of bits.

Quoted 14 lines

Documentation/serial/serialsim.rst line 136

Quoted 10 lines
Documentation/serial/serialsim.rst new file +9 -0
+    a 16-bit mask telling which values to set, bits 0-15 are the
+    actual values.  Settable values are TIOCM_CAR, TIOCM_CTS,
+    TIOCM_DSR, and TIOC_RNG.  If NULLMODEM is set to true, then only
+    TIOC_RNG is settable.  The DTR and RTS lines are not here, you can
+    set them through the normal interface.
+
+TIOCSERSREMERR
+    Send an error or errors on the next sent byte.  arg is
+    a bitwise OR of (1 << TTY_xxx).  Allowed errors are TTY_BREAK,

is this better: (or I don't understand?)
a bitwise OR of (1 << TTY_xxx) and one (or more of) the allowed error flags TTY_BREAK, TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.

Quoted 6 lines

Well, not really. But what I wrote isn't great, so, how about:

Send an error or errors on the next sent byte. arg is a bitwise OR of (1 << TTY_BREAK), (1 << TTY_FRAME), (1 << TTY_PARITY), and (1 << TTY_OVERRUN). If none of those are set, then no error is sent.

I see. Thanks.

Quoted 23 lines

Thanks,

-corey

Quoted 19 lines
Quoted 4 lines
Documentation/serial/serialsim.rst new file +3 -0
+    TTY_FRAME, TTY_PARITY, and TTY_OVERRUN.
+
+TIOCSERGREMTERMIOS

Documentation/serial/serialsim.rst line 149

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+    Return the termios structure for the other side of the pipe.
+    arg is a pointer to a standard termios struct.
+
+TIOCSERGREMRS485
+    Return the remote RS485 settings, arg is a pointer to a struct
+    serial_rs485.
+
+Note that unlike the sysfs interface, these ioctls affect the other
+end.  So setting nullmodem on the ttyPipeB0 interface sets whether
+the DTR/RTS lines on ttyPipeB0 affect ttyPipeA0.

cheers.

Signature · 1 line
~Randy
Signature · 1 line
~Randy
#4 lore raw

Dumb question: this is basically a pty on steroids. Wouldn't this be better done by enhancing the pty devices?

	-0hpa
#5 lore raw
Quoted 2 lines · H. Peter Anvin

Dumb question: this is basically a pty on steroids. Wouldn't this be better done by enhancing the pty devices?

I did look at that, but it would be pretty invasive to pty. There's no modem control stuff, none of the other special serial ioctls. And the locking in this driver is fairly strange because you have two serial ports looking at each other's data for modem control. But that might not be a big deal.

Adding the speed simulation to ptys would also be really strange. That's not a deal-breaker, I suppose, but it's not much of a serial port simulation without it.

-corey

#6 lore raw
Quoted 4 lines · Corey Minyard
Quoted 2 lines · H. Peter Anvin

Dumb question: this is basically a pty on steroids. Wouldn't this be better done by enhancing the pty devices?

I proposed doing that several years ago, and offered to start working on it if there was a decent chance it would be accepted into the tree. I got no response.

Quoted 9 lines

I did look at that, but it would be pretty invasive to pty. There's no modem control stuff, none of the other special serial ioctls. And the locking in this driver is fairly strange because you have two serial ports looking at each other's data for modem control. But that might not be a big deal.

Adding the speed simulation to ptys would also be really strange. That's not a deal-breaker, I suppose, but it's not much of a serial port simulation without it.

My goal wasn't really to simulate two serial ports with a null-mode cable in-between, so the speed simluaiton wasn't on my list. my goal was to provide a way to implement a serial port in userspace by attaching an application to the master end of a pty.

The pty(7) man page states "The slave end of the pseudoterminal provides an interface that behaves exactly like a classical terminal." But, we all know that's a pretty big lie: the pty slave end implements only a small subset of a "classical termial" device, and there are all sorts of applications that expect to talk to a serial port which fail miserably when connected to a pty.

Signature · 3 lines
Grant Edwards               grant.b.edwards        Yow! I had a lease on an
                                  at               OEDIPUS COMPLEX back in
                              gmail.com            '81 ...
#7 lore raw
In reply to [email protected]
Quoted 6 lines · On Tue, Mar 05, 2019 at 11:12:31AM -0600, [email protected] wrote:

From: Corey Minyard <[email protected]>

This creates simulated serial ports, both as echo devices and pipe devices. The driver reasonably approximates the serial port speed and simulates some modem control lines. It allows error injection and direct control of modem control lines.

I like this, thanks for doing it!

Minor nits below:

Quoted 62 lines
Documentation/serial/serialsim.rst new file +62 -0
+You can modify null modem and control the lines individually
+through an interface in /sys/class/tty/ttyECHO<n>/ctrl,
+/sys/class/tty/ttyPipeA<n>/ctrl, and
+/sys/class/tty/ttyPipeB<n>/ctrl.  The following may be written to
+those files:
+
+[+-]nullmodem
+    enable/disable null modem
+
+[+-]cd
+    enable/disable Carrier Detect (no effect if +nullmodem)
+
+[+-]dsr
+    enable/disable Data Set Ready (no effect if +nullmodem)
+
+[+-]cts
+    enable/disable Clear To Send (no effect if +nullmodem)
+
+[+-]ring
+    enable/disable Ring
+
+frame
+    inject a frame error on the next byte
+
+parity
+    inject a parity error on the next byte
+
+overrun
+    inject an overrun error on the next byte
+
+The contents of the above files has the following format:
+
+tty[Echo|PipeA|PipeB]<n>
+    <mctrl values>
+
+where <mctrl values> is the modem control values above (not frame,
+parity, or overrun) with the following added:
+
+[+-]dtr
+    value of the Data Terminal Ready
+
+[+-]rts
+    value of the Request To Send
+
+The above values are not settable through this interface, they are
+set through the serial port interface itself.
+
+So, for instance, ttyEcho0 comes up in the following state::
+
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem -cd -dsr -cts -ring -dtr -rts
+
+If something connects, it will become::
+
+   ttyEcho0: +nullmodem +cd +dsr +cts -ring +dtr +rts
+
+To enable ring::
+
+   # echo "+ring" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: +nullmodem +cd +dsr +cts +ring +dtr +rts
+

Documentation/serial/serialsim.rst line 117

Quoted 10 lines
Documentation/serial/serialsim.rst new file +10 -0
+Now disable NULL modem and the CD line::
+
+   # echo "-nullmodem -cd" >/sys/class/tty/ttyEcho0/ctrl
+   # cat /sys/class/tty/ttyEcho0/ctrl
+   ttyEcho0: -nullmodem -cd -dsr -cts +ring -dtr -rts
+
+Note that these settings are for the side you are modifying.  So if
+you set nullmodem on ttyPipeA0, that controls whether the DTR/RTS
+lines from ttyPipeB0 affect ttyPipeA0.  It doesn't affect ttyPipeB's
+modem control lines.

All of the sysfs stuff needs to also go in Documentation/ABI to describe your new files.

drivers/tty/serial/Kconfig endmenu · line 1570

Quoted 3 lines
drivers/tty/serial/Kconfig +3 -0
+config SERIAL_SIMULATOR
+	tristate "Serial port simulator"
+	default n

n is the default, no need to say it again here.

drivers/tty/serial/serialsim.c line 6

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +6 -0
@@ -0,0 +1,1101 @@
+// SPDX-License-Identifier: GPL-2.0+
+
+/*
+ * serialsim - Emulate a serial device in a loopback and/or pipe
+ *
+ * See the docs for this for more details.

Pointer to the docs?

And no copyright notice? I don't object to it, but it is usually nice to see.

Quoted 15 lines
drivers/tty/serial/serialsim.c new file +15 -0
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/ioport.h>
+#include <linux/init.h>
+#include <linux/serial.h>
+#include <linux/tty.h>
+#include <linux/tty_flip.h>
+#include <linux/device.h>
+#include <linux/serial_core.h>
+#include <linux/kthread.h>
+#include <linux/hrtimer.h>
+#include <linux/slab.h>

drivers/tty/serial/serialsim.c line 31

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+#include <linux/spinlock.h>
+#include <linux/ctype.h>
+#include <linux/string.h>
+#include <linux/uaccess.h>
+
+#include <linux/serialsim.h>
+
+#define PORT_SERIALECHO 72549
+#define PORT_SERIALPIPEA 72550
+#define PORT_SERIALPIPEB 72551

tabs?

Quoted 57 lines
drivers/tty/serial/serialsim.c new file +57 -0
+
+#ifdef CONFIG_HIGH_RES_TIMERS
+#define SERIALSIM_WAKES_PER_SEC	1000
+#else
+#define SERIALSIM_WAKES_PER_SEC	HZ
+#endif
+
+#define SERIALSIM_XBUFSIZE	32
+
+/* For things to send on the line, in flags field. */
+#define DO_FRAME_ERR		(1 << TTY_FRAME)
+#define DO_PARITY_ERR		(1 << TTY_PARITY)
+#define DO_OVERRUN_ERR		(1 << TTY_OVERRUN)
+#define DO_BREAK		(1 << TTY_BREAK)
+#define FLAGS_MASK (DO_FRAME_ERR | DO_PARITY_ERR | DO_OVERRUN_ERR | DO_BREAK)
+
+struct serialsim_intf {
+	struct uart_port port;
+
+	/*
+	 * This is my transmit buffer, my thread picks this up and
+	 * injects them into the other port's uart.
+	 */
+	unsigned char xmitbuf[SERIALSIM_XBUFSIZE];
+	struct circ_buf buf;
+
+	/* Error flags to send. */
+	bool break_reported;
+	unsigned int flags;
+
+	/* Modem state. */
+	unsigned int mctrl;
+	bool do_null_modem;
+	spinlock_t mctrl_lock;
+	struct tasklet_struct mctrl_tasklet;
+
+	/* My transmitter is enabled. */
+	bool tx_enabled;
+
+	/* I can receive characters. */
+	bool rx_enabled;
+
+	/* Is the port registered with the uart driver? */
+	bool registered;
+
+	/*
+	 * The serial echo port on the other side of this pipe (or points
+	 * to myself in loopback mode.
+	 */
+	struct serialsim_intf *ointf;
+
+	unsigned int div;
+	unsigned int bytes_per_interval;
+	unsigned int per_interval_residual;
+
+	struct ktermios termios;
+

drivers/tty/serial/serialsim.c line 98

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+	const char *threadname;
+	struct task_struct *thread;
+
+	struct serial_rs485 rs485;
+};
+
+#define circ_sbuf_space(buf) CIRC_SPACE((buf)->head, (buf)->tail, \
+					SERIALSIM_XBUFSIZE)
+#define circ_sbuf_empty(buf) ((buf)->head == (buf)->tail)
+#define circ_sbuf_next(idx) (((idx) + 1) % SERIALSIM_XBUFSIZE)

Don't we have a ring buffer (or 3) structure already? Did you create another one?

drivers/tty/serial/serialsim.c line 381

Quoted 12 lines
drivers/tty/serial/serialsim.c new file +12 -0
+static void serialsim_thread_delay(void)
+{
+#ifdef CONFIG_HIGH_RES_TIMERS
+	ktime_t timeout;
+
+	timeout = ns_to_ktime(1000000000 / SERIALSIM_WAKES_PER_SEC);
+	set_current_state(TASK_INTERRUPTIBLE);
+	schedule_hrtimeout(&timeout, HRTIMER_MODE_REL);
+#else
+	schedule_timeout_interruptible(1);
+#endif
+}

Why do you care about hires timers here? Why not always just sleep to slow things down?

drivers/tty/serial/serialsim.c line 392

Quoted 11 lines
drivers/tty/serial/serialsim.c new file +11 -0
+
+static int serialsim_thread(void *data)
+{
+	struct serialsim_intf *intf = data;
+	struct serialsim_intf *ointf = intf->ointf;
+	struct uart_port *port = &intf->port;
+	struct uart_port *oport = &ointf->port;
+	struct circ_buf *tbuf = &intf->buf;
+	unsigned int residual = 0;
+
+	while (!kthread_should_stop()) {

Aren't we trying to get away from creating new kthreads?

Can you just use a workqueue entry?

drivers/tty/serial/serialsim.c line 689

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +9 -0
+static unsigned int nr_echo_ports = 4;
+module_param(nr_echo_ports, uint, 0444);
+MODULE_PARM_DESC(nr_echo_ports,
+		 "The number of echo ports to create.  Defaults to 4");
+
+static unsigned int nr_pipe_ports = 4;
+module_param(nr_pipe_ports, uint, 0444);
+MODULE_PARM_DESC(nr_pipe_ports,
+		 "The number of pipe ports to create.  Defaults to 4");

No way to just do this with configfs and not worry about module params?

drivers/tty/serial/serialsim.c line 703

Quoted 13 lines
drivers/tty/serial/serialsim.c new file +13 -0
+static char *gettok(char **s)
+{
+	char *t = skip_spaces(*s);
+	char *p = t;
+
+	while (*p && !isspace(*p))
+		p++;
+	if (*p)
+		*p++ = '\0';
+	*s = p;
+
+	return t;
+}

We don't have this already in the tree?

drivers/tty/serial/serialsim.c line 708

Quoted 4 lines
drivers/tty/serial/serialsim.c new file +4 -0
+static bool tokeq(const char *t, const char *m)
+{
+	return strcmp(t, m) == 0;
+}

same here.

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +9 -0
+
+static unsigned int parse_modem_line(char op, unsigned int flag,
+				     unsigned int *mctrl)
+{
+	if (op == '+')
+		*mctrl |= flag;
+	else
+		*mctrl &= ~flag;
+	return flag;

drivers/tty/serial/serialsim.c line 727

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+}
+
+static void serialsim_ctrl_append(char **val, int *left, char *n, bool enabled)
+{
+	int count;
+
+	count = snprintf(*val, *left, " %c%s", enabled ? '+' : '-', n);
+	*left -= count;
+	*val += count;
+}

sysfs files really should only be "one value per file", so this api is troubling :(

Quoted 4 lines
drivers/tty/serial/serialsim.c new file +4 -0
+static ssize_t serialsim_ctrl_read(struct device *dev,
+				   struct device_attribute *attr,
+				   char *buf)
+{

drivers/tty/serial/serialsim.c line 742

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+	struct tty_port *tport = dev_get_drvdata(dev);
+	struct uart_state *state = container_of(tport, struct uart_state, port);
+	struct uart_port *port = state->uart_port;
+	struct serialsim_intf *intf = serialsim_port_to_intf(port);
+	unsigned int mctrl = intf->mctrl;
+	char *val = buf;
+	int left = PAGE_SIZE;
+	int count;
+
+	count = snprintf(val, left, "%s:", dev->kobj.name);

dev_name()?

And is it really needed? It's already known as this is the sysfs file for that device. Don't make userspace parse it away.

Quoted 74 lines
drivers/tty/serial/serialsim.c new file +74 -0
+	val += count;
+	left -= count;
+	serialsim_ctrl_append(&val, &left, "nullmodem", intf->do_null_modem);
+	serialsim_ctrl_append(&val, &left, "cd", mctrl & TIOCM_CAR);
+	serialsim_ctrl_append(&val, &left, "dsr", mctrl & TIOCM_DSR);
+	serialsim_ctrl_append(&val, &left, "cts", mctrl & TIOCM_CTS);
+	serialsim_ctrl_append(&val, &left, "ring", mctrl & TIOCM_RNG);
+	serialsim_ctrl_append(&val, &left, "dtr", mctrl & TIOCM_DTR);
+	serialsim_ctrl_append(&val, &left, "rts", mctrl & TIOCM_RTS);
+	*val++ = '\n';
+
+	return val - buf;
+}
+
+static ssize_t serialsim_ctrl_write(struct device *dev,
+				    struct device_attribute *attr,
+				    const char *val, size_t count)
+{
+	struct tty_port *tport = dev_get_drvdata(dev);
+	struct uart_state *state = container_of(tport, struct uart_state, port);
+	struct uart_port *port = state->uart_port;
+	struct serialsim_intf *intf = serialsim_port_to_intf(port);
+	char *str = kstrndup(val, count, GFP_KERNEL);
+	char *p, *s = str;
+	int rv = count;
+	unsigned int flags = 0;
+	unsigned int nullmodem = 0;
+	unsigned int mctrl_mask = 0, mctrl = 0;
+
+	if (!str)
+		return -ENOMEM;
+
+	p = gettok(&s);
+	while (*p) {
+		char op = '\0';
+		int err = 0;
+
+		switch (*p) {
+		case '+':
+		case '-':
+			op = *p++;
+			break;
+		default:
+			break;
+		}
+
+		if (tokeq(p, "frame"))
+			flags |= DO_FRAME_ERR;
+		else if (tokeq(p, "parity"))
+			flags |= DO_PARITY_ERR;
+		else if (tokeq(p, "overrun"))
+			flags |= DO_OVERRUN_ERR;
+		else if (tokeq(p, "nullmodem"))
+			nullmodem = op;
+		else if (tokeq(p, "dsr"))
+			mctrl_mask |= parse_modem_line(op, TIOCM_DSR, &mctrl);
+		else if (tokeq(p, "cts"))
+			mctrl_mask |= parse_modem_line(op, TIOCM_CTS, &mctrl);
+		else if (tokeq(p, "cd"))
+			mctrl_mask |= parse_modem_line(op, TIOCM_CAR, &mctrl);
+		else if (tokeq(p, "ring"))
+			mctrl_mask |= parse_modem_line(op, TIOCM_RNG, &mctrl);
+		else
+			err = -EINVAL;
+
+		if (err) {
+			rv = err;
+			goto out;
+		}
+		p = gettok(&s);
+	}
+
+	if (flags)
+		serialsim_set_flags(intf, flags);

drivers/tty/serial/serialsim.c line 826

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+	if (nullmodem)
+		serialsim_set_null_modem(intf, nullmodem == '+');
+	if (mctrl_mask)
+		serialsim_set_modem_lines(intf, mctrl_mask, mctrl);
+
+out:
+	kfree(str);
+
+	return rv;
+}

This worries me, parsing sysfs files is ripe for problems. configfs might be better here.

drivers/tty/serial/serialsim.c line 828

Quoted 2 lines
drivers/tty/serial/serialsim.c new file +2 -0
+
+static DEVICE_ATTR(ctrl, 0660, serialsim_ctrl_read, serialsim_ctrl_write);

DEVICE_ATTR_RW()?

drivers/tty/serial/serialsim.c line 837

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +9 -0
+
+static struct attribute *serialsim_dev_attrs[] = {
+	&dev_attr_ctrl.attr,
+	NULL,
+};
+
+static struct attribute_group serialsim_dev_attr_group = {
+	.attrs = serialsim_dev_attrs,
+};

ATTRIBUTE_GROUPS()?

drivers/tty/serial/serialsim.c line 917

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+static int __init serialsim_init(void)
+{
+	unsigned int i;
+	int rv;
+
+	serialecho_ports = kcalloc(nr_echo_ports,
+				   sizeof(*serialecho_ports),
+				   GFP_KERNEL);
+	if (!serialecho_ports) {
+		pr_err("serialsim: Unable to allocate echo ports.\n");

No need to print error for memory failure.

drivers/tty/serial/serialsim.c line 286

Quoted 1 line
drivers/tty/serial/serialsim.c new file +1 -0
+		rv = ENOMEM;

-ENOMEM?

drivers/tty/serial/serialsim.c line 927

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +9 -0
+		goto out;
+	}
+
+	serialpipe_ports = kcalloc(nr_pipe_ports * 2,
+				   sizeof(*serialpipe_ports),
+				   GFP_KERNEL);
+	if (!serialpipe_ports) {
+		kfree(serialecho_ports);
+		pr_err("serialsim: Unable to allocate pipe ports.\n");

Same here.

drivers/tty/serial/serialsim.c line 296

Quoted 1 line
drivers/tty/serial/serialsim.c new file +1 -0
+		rv = ENOMEM;

-ENOMEM?

drivers/tty/serial/serialsim.c line 297

Quoted 1 line
drivers/tty/serial/serialsim.c new file +1 -0
+		goto out;

Shouldn't out clean up stuff?

drivers/tty/serial/serialsim.c line 938

Quoted 9 lines
drivers/tty/serial/serialsim.c new file +9 -0
+	}
+
+	serialecho_driver.nr = nr_echo_ports;
+	rv = uart_register_driver(&serialecho_driver);
+	if (rv) {
+		kfree(serialecho_ports);
+		kfree(serialpipe_ports);
+		pr_err("serialsim: Unable to register driver.\n");
+		goto out;

Again, out should clean up stuff for an error. Will make the code below simpler.

Quoted 107 lines
drivers/tty/serial/serialsim.c new file +107 -0
+	}
+
+	serialpipea_driver.nr = nr_pipe_ports;
+	rv = uart_register_driver(&serialpipea_driver);
+	if (rv) {
+		uart_unregister_driver(&serialecho_driver);
+		kfree(serialecho_ports);
+		kfree(serialpipe_ports);
+		pr_err("serialsim: Unable to register driver.\n");
+		goto out;
+	}
+
+	serialpipeb_driver.nr = nr_pipe_ports;
+	rv = uart_register_driver(&serialpipeb_driver);
+	if (rv) {
+		uart_unregister_driver(&serialpipea_driver);
+		uart_unregister_driver(&serialecho_driver);
+		kfree(serialecho_ports);
+		kfree(serialpipe_ports);
+		pr_err("serialsim: Unable to register driver.\n");
+		goto out;
+	}
+
+	for (i = 0; i < nr_echo_ports; i++) {
+		struct serialsim_intf *intf = &serialecho_ports[i];
+		struct uart_port *port = &intf->port;
+
+		intf->buf.buf = intf->xmitbuf;
+		intf->ointf = intf;
+		intf->threadname = "serialecho";
+		intf->do_null_modem = true;
+		spin_lock_init(&intf->mctrl_lock);
+		tasklet_init(&intf->mctrl_tasklet, mctrl_tasklet, (long) intf);
+		/* Won't configure without some I/O or mem address set. */
+		port->iobase = 1;
+		port->line = i;
+		port->flags = UPF_BOOT_AUTOCONF;
+		port->ops = &serialecho_ops;
+		spin_lock_init(&port->lock);
+		port->attr_group = &serialsim_dev_attr_group;
+		rv = uart_add_one_port(&serialecho_driver, port);
+		if (rv)
+			pr_err("serialsim: Unable to add uart port %d: %d\n",
+			       i, rv);
+		else
+			intf->registered = true;
+	}
+
+	for (i = 0; i < nr_pipe_ports * 2; i += 2) {
+		struct serialsim_intf *intfa = &serialpipe_ports[i];
+		struct serialsim_intf *intfb = &serialpipe_ports[i + 1];
+		struct uart_port *porta = &intfa->port;
+		struct uart_port *portb = &intfb->port;
+
+		intfa->buf.buf = intfa->xmitbuf;
+		intfb->buf.buf = intfb->xmitbuf;
+		intfa->ointf = intfb;
+		intfb->ointf = intfa;
+		intfa->threadname = "serialpipea";
+		intfb->threadname = "serialpipeb";
+		spin_lock_init(&intfa->mctrl_lock);
+		spin_lock_init(&intfb->mctrl_lock);
+		tasklet_init(&intfa->mctrl_tasklet, mctrl_tasklet,
+			     (long) intfa);
+		tasklet_init(&intfb->mctrl_tasklet, mctrl_tasklet,
+			     (long) intfb);
+
+		/* Won't configure without some I/O or mem address set. */
+		porta->iobase = 1;
+		porta->line = i / 2;
+		porta->flags = UPF_BOOT_AUTOCONF;
+		porta->ops = &serialpipea_ops;
+		spin_lock_init(&porta->lock);
+		porta->attr_group = &serialsim_dev_attr_group;
+		porta->rs485_config = serialsim_rs485;
+
+		/*
+		 * uart_add_one_port() does an mctrl operation, which will
+		 * claim the other port's lock.  So everything needs to be
+		 * full initialized, and we need null modem off until we
+		 * get things added.
+		 */
+		portb->iobase = 1;
+		portb->line = i / 2;
+		portb->flags = UPF_BOOT_AUTOCONF;
+		portb->ops = &serialpipeb_ops;
+		portb->attr_group = &serialsim_dev_attr_group;
+		spin_lock_init(&portb->lock);
+		portb->rs485_config = serialsim_rs485;
+
+		rv = uart_add_one_port(&serialpipea_driver, porta);
+		if (rv) {
+			pr_err("serialsim: Unable to add uart pipe a port %d: %d\n",
+			       i, rv);
+			continue;
+		} else {
+			intfa->registered = true;
+		}
+
+		rv = uart_add_one_port(&serialpipeb_driver, portb);
+		if (rv) {
+			pr_err("serialsim: Unable to add uart pipe b port %d: %d\n",
+			       i, rv);
+			intfa->registered = false;
+			uart_remove_one_port(&serialpipea_driver, porta);
+		} else {
+			intfb->registered = true;

drivers/tty/serial/serialsim.c line 1055

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+		}
+
+		if (intfa->registered && intfb->registered) {
+			serialsim_set_null_modem(intfa, true);
+			serialsim_set_null_modem(intfb, true);
+		}
+	}
+	rv = 0;
+
+	pr_info("serialsim ready\n");

Don't be noisy for when all is good. Drivers should be quiet.

Quoted 23 lines
include/uapi/linux/serialsim.h new file +20 -0
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+
+/*
+ * serialsim - Emulate a serial device in a loopback and/or pipe
+ */
+
+/*
+ * TTY IOCTLs for controlling the modem control and for error injection.
+ * See drivers/tty/serial/serialsim.c and Documentation/serial/serialsim.rst
+ * for details.
+ */
+
+#ifndef LINUX_SERIALSIM_H
+#define LINUX_SERIALSIM_H
+
+#include <linux/ioctl.h>
+#include <asm/termbits.h>
+
+#define TIOCSERSREMNULLMODEM	0x54e4
+#define TIOCSERSREMMCTRL	0x54e5

include/uapi/linux/serialsim.h line 30

Quoted 10 lines
include/uapi/linux/serialsim.h new file +10 -0
+#define TIOCSERSREMERR		0x54e6
+#ifdef TCGETS2
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios2)
+#else
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios)
+#endif
+#define TIOCSERGREMNULLMODEM	_IOR('T', 0xe8, int)
+#define TIOCSERGREMMCTRL	_IOR('T', 0xe9, unsigned int)
+#define TIOCSERGREMERR		_IOR('T', 0xea, unsigned int)
+#define TIOCSERGREMRS485	_IOR('T', 0xeb, struct serial_rs485)

Wait, don't we have ioctls for these things in the tty layer already? WHy add new ones?

thanks,

greg k-h

#8 lore raw
Quoted 9 lines · Greg Kroah-Hartman
Quoted 6 lines · On Tue, Mar 05, 2019 at 11:12:31AM -0600, [email protected] wrote:

From: Corey Minyard <[email protected]>

This creates simulated serial ports, both as echo devices and pipe devices. The driver reasonably approximates the serial port speed and simulates some modem control lines. It allows error injection and direct control of modem control lines.

I like this, thanks for doing it!

Hopefully this is useful for others. I guess you could test the kernel serial code with it, too.

Quoted 2 lines

Minor nits below:

Deleted and made the changes I agreed with or understood. Other comment below...

Quoted 8 lines

drivers/tty/serial/serialsim.c line 98

Quoted 5 lines
drivers/tty/serial/serialsim.c new file +5 -0
+
+#define circ_sbuf_space(buf) CIRC_SPACE((buf)->head, (buf)->tail, \
+					SERIALSIM_XBUFSIZE)
+#define circ_sbuf_empty(buf) ((buf)->head == (buf)->tail)
+#define circ_sbuf_next(idx) (((idx) + 1) % SERIALSIM_XBUFSIZE)

Don't we have a ring buffer (or 3) structure already? Did you create another one?

I'm using circ_buf, same as the main serial code, these are just helpers. Is there something else I can use? The only ring buffer I saw that was named that was for tracing, and that really wouldn't work.

Quoted 16 lines

drivers/tty/serial/serialsim.c line 381

Quoted 13 lines
drivers/tty/serial/serialsim.c new file +12 -0
+static void serialsim_thread_delay(void)
+{
+#ifdef CONFIG_HIGH_RES_TIMERS
+	ktime_t timeout;
+
+	timeout = ns_to_ktime(1000000000 / SERIALSIM_WAKES_PER_SEC);
+	set_current_state(TASK_INTERRUPTIBLE);
+	schedule_hrtimeout(&timeout, HRTIMER_MODE_REL);
+#else
+	schedule_timeout_interruptible(1);
+#endif
+}

Why do you care about hires timers here? Why not always just sleep to slow things down?

It makes things smoother, otherwise you get no data for a while then big bursts of data at higher speeds.

Quoted 15 lines

drivers/tty/serial/serialsim.c line 392

Quoted 12 lines
drivers/tty/serial/serialsim.c new file +11 -0
+
+static int serialsim_thread(void *data)
+{
+	struct serialsim_intf *intf = data;
+	struct serialsim_intf *ointf = intf->ointf;
+	struct uart_port *port = &intf->port;
+	struct uart_port *oport = &ointf->port;
+	struct circ_buf *tbuf = &intf->buf;
+	unsigned int residual = 0;
+
+	while (!kthread_should_stop()) {

Aren't we trying to get away from creating new kthreads?

Quoted 1 line

Can you just use a workqueue entry?

The purpose of this is to wait short periods of time and copy characters from one serial port to the other. I'm not sure how a work queue would do that. I could drive it from timers, would that be better? I don't need something that runs in thread context.

Either way, it's not a terribly efficient mechanism, but that wasn't a big concern for this, I don't think.

Quoted 12 lines

drivers/tty/serial/serialsim.c line 689

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +9 -0
+static unsigned int nr_echo_ports = 4;
+module_param(nr_echo_ports, uint, 0444);
+MODULE_PARM_DESC(nr_echo_ports,
+		 "The number of echo ports to create.  Defaults to 4");
+
+static unsigned int nr_pipe_ports = 4;
+module_param(nr_pipe_ports, uint, 0444);
+MODULE_PARM_DESC(nr_pipe_ports,
+		 "The number of pipe ports to create.  Defaults to 4");

No way to just do this with configfs and not worry about module params?

I'll look. Module parameters seem a lot simpler, though. I could also just make it not configurable.

Quoted 16 lines
Quoted 4 lines
drivers/tty/serial/serialsim.c new file +3 -0
+static char *gettok(char **s)
+{
+	char *t = skip_spaces(*s);

drivers/tty/serial/serialsim.c line 703

Quoted 10 lines
drivers/tty/serial/serialsim.c new file +10 -0
+	char *p = t;
+
+	while (*p && !isspace(*p))
+		p++;
+	if (*p)
+		*p++ = '\0';
+	*s = p;
+
+	return t;
+}

We don't have this already in the tree?

There is strsep(), but there's no way to do isspace() on it. I don't see a set of space characters anyplace.

Quoted 7 lines

drivers/tty/serial/serialsim.c line 708

Quoted 5 lines
drivers/tty/serial/serialsim.c new file +4 -0
+static bool tokeq(const char *t, const char *m)
+{
+	return strcmp(t, m) == 0;
+}

same here.

There is sysfs_streq(), but it's not quite the same. I didn't see anything else.

Quoted 15 lines
Quoted 15 lines
drivers/tty/serial/serialsim.c new file +14 -0
+
+static unsigned int parse_modem_line(char op, unsigned int flag,
+				     unsigned int *mctrl)
+{
+	if (op == '+')
+		*mctrl |= flag;
+	else
+		*mctrl &= ~flag;
+	return flag;
+}
+
+static void serialsim_ctrl_append(char **val, int *left, char *n, bool enabled)
+{
+	int count;
Quoted 8 lines

drivers/tty/serial/serialsim.c line 727

Quoted 5 lines
drivers/tty/serial/serialsim.c new file +5 -0
+
+	count = snprintf(*val, *left, " %c%s", enabled ? '+' : '-', n);
+	*left -= count;
+	*val += count;
+}

sysfs files really should only be "one value per file", so this api is troubling :(

Yeah, it is. I wanted something that was easy for scripts to parse. More on this below...

Quoted 3 lines

This worries me, parsing sysfs files is ripe for problems. configfs might be better here.

Maybe so, but wouldn't you still have to do parsing?

I could separate these out into individual items (dtr, dsr, etc.). You wouldn't be able to get an atomic view of the modem lines, though.

On the whole sysfs thing, I'm not currently using the sysfs interface for my testing, but I thought I would use it from scripts. It would still be better for scripts, but perhaps it's better to just pull the whole sysfs part.

Quoted 5 lines

drivers/tty/serial/serialsim.c line 828

Quoted 3 lines
drivers/tty/serial/serialsim.c new file +2 -0
+
+static DEVICE_ATTR(ctrl, 0660, serialsim_ctrl_read, serialsim_ctrl_write);

DEVICE_ATTR_RW()?

Doesn't take show or store functions.

Quoted 8 lines
Quoted 8 lines
include/uapi/linux/serialsim.h new file +8 -0
+
+#define TIOCSERSREMNULLMODEM	0x54e4
+#define TIOCSERSREMMCTRL	0x54e5
+#define TIOCSERSREMERR		0x54e6
+#ifdef TCGETS2
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios2)
+#else
+#define TIOCSERGREMTERMIOS	_IOR('T', 0xe7, struct termios)
Quoted 8 lines

include/uapi/linux/serialsim.h line 30

Quoted 5 lines
include/uapi/linux/serialsim.h new file +5 -0
+#endif
+#define TIOCSERGREMNULLMODEM	_IOR('T', 0xe8, int)
+#define TIOCSERGREMMCTRL	_IOR('T', 0xe9, unsigned int)
+#define TIOCSERGREMERR		_IOR('T', 0xea, unsigned int)
+#define TIOCSERGREMRS485	_IOR('T', 0xeb, struct serial_rs485)

Wait, don't we have ioctls for these things in the tty layer already? WHy add new ones?

These are for getting and setting the remote end's modem control lines. I'm not sure how I could do that with the same ioctls.

-corey