README
======
Contents
========
o Buffering Notes
- Hardware Flow Control
- RX Buffer Size
- Buffer Recommendations
o Using NuttX ZModem with a Linux Host
- Sending Files from the Target to the Linux Host PC
- Receiving Files on the Target from the Linux Host PC
o Building the ZModem Tools to Run Under Linux
o Status
Buffering Notes
===============
Hardware Flow Control
---------------------
Hardware flow control must be enabled in serial drivers in order to
prevent data overrun. However, in the most NuttX serial drivers, hardware
flow control only protects the hardware RX FIFO: Data will not be lost in
the hardware FIFO but can still be lost when it is taken from the FIFO.
We can still overflow the serial driver's RX buffer even with hardware
flow control enabled! That is probably a bug. But the workaround solution
that I have used is to use lower data rates and a large serial driver RX
buffer.
Those measures should be unnecessary if buffering and hardware flow
control are set up and working correctly.
Software Flow Control
---------------------
The ZModem protocol has XON/XOFF flow control built into it. The protocol
permits XON or XOFF characters placed at certain parts of messages. If
software flow control is enabled on the receiving end it will consume the
XONs and XOFFs. Otherwise they will be ignored in the data by the ZModem
logic.
NuttX, however, does not implement XON/XOFF flow control so these do
nothing. On NuttX you will have to use hardware flow control in most cases.
The XON/XOFF controls built into ZModem could be used if you enabled
software flow control in the host. But that would only work in one
direction: If would prevent the host from overrunning the the target Rx
buffering. So you should be able to do host-to-target software flow
control. But there would still be no target-to-host flow control. That
might not be an issue because the host is usually so much faster than
that target.
RX Buffer Size
--------------
The ZModem protocol supports a message that informs the file sender of
the maximum size of data that you can buffer (ZRINIT). However, my
experience is that the Linux sz ignores this setting and always sends file
data at the maximum size (1024) no matter what size of buffer you report.
That is unfortunate because that, combined with the possibilities of data
overrun mean that you must use quite large buffering for ZModem file
receipt to be reliable (none of these issues effect sending of files).
Buffer Recommendations
----------------------
Based on the limitations of NuttX hardware flow control and of the Linux
sz behavior, I have been testing with the following configuration
(assuming UART1 is the ZModem device):
1) This setting determines that maximum size of a data packet frame:
CONFIG_SYSTEM_ZMODEM_PKTBUFSIZE=1024
2) Input Buffering. If the input buffering is set to a full frame, then
data overflow is less likely.
CONFIG_UART1_RXBUFSIZE=1024
3) With a larger driver input buffer, the ZModem receive I/O buffer can be
smaller:
CONFIG_SYSTEM_ZMODEM_RCVBUFSIZE=256
4) Output buffering. Overrun cannot occur on output (on the NuttX side)
so there is no need to be so careful:
CONFIG_SYSTEM_ZMODEM_SNDBUFSIZE=512
CONFIG_UART1_TXBUFSIZE=256
Using NuttX ZModem with a Linux Host
====================================
Sending Files from the Target to the Linux Host PC
--------------------------------------------------
The NuttX ZModem commands have been verified against the rzsz programs
running on a Linux PC. To send a file to the PC, first make sure that
the serial port is configured to work with the board (Assuming you are
using 9600 baud for the data transfers -- high rates may result in data
overruns):
$ sudo stty -F /dev/ttyS0 9600 # Select 9600 BAUD
$ sudo stty -F /dev/ttyS0 crtscts # Enables CTS/RTS handshaking *
$ sudo stty -F /dev/ttyS0 raw # Puts the TTY in raw mode
$ sudo stty -F /dev/ttyS0 # Show the TTY configuration
* Only if hardware flow control is enabled.
Start rz on the Linux host (using /dev/ttyS0 as an example):
$ sudo rz </dev/ttyS0 >/dev/ttyS0
You can add the rz -v option multiple times, each increases the level
of debug output. If you want to capture the Linux rz output, then
re-direct stderr to a log file by adding 2>rz.log to the end of the
rz command.
NOTE: The NuttX ZModem does sends rz\n when it starts in compliance with
the ZModem specification. On Linux this, however, seems to start some
other, incompatible version of rz. You need to start rz manually to
make sure that the correct version is selected. You can tell when this
evil rz/sz has inserted itself because you will see the '^' (0x5e)
character replacing the standard ZModem ZDLE character (0x19) in the
binary data stream.
If you don't have the rz command on your Linux box, the package to
install rzsz (or possibily lrzsz).
Then on the target (using /dev/ttyS1 as an example).
> sz -d /dev/ttyS1 <filename>
Where filename is the full path to the file to send (i.e., it begins
with the '/' character). /dev/ttyS1 or whatever device you select
*MUST* support Hardware flow control in order to throttle therates of
data transfer to fit within the allocated buffers.
Receiving Files on the Target from the Linux Host PC
----------------------------------------------------
NOTE: There are issues with using the Linux sz command with the NuttX
rz command. See "Status" below. It is recommended that you use the
NuttX sz command on Linux as described in the next paragraph.
To send a file to the target, first make sure that the serial port on the
host is configured to work with the board (Assuming that you are using
9600 baud for the data transfers -- high rates may result in data
overruns):
$ sudo stty -F /dev/ttyS0 9600 # Select 9600 (or other) BAUD
$ sudo stty -F /dev/ttyS0 crtscts # Enables CTS/RTS handshaking *
$ sudo stty -F /dev/ttyS0 raw # Puts the TTY in raw mode
$ sudo stty -F /dev/ttyS0 # Show the TTY configuration
* Only is hardware flow control is enabled.
Start rz on the on the target. Here, in this example, we are using
/dev/ttyS1 to perform the transfer
nsh> rz -d /dev/ttyS1
/dev/ttyS1 or whatever device you select *MUST* support Hardware flow
control in order to throttle therates of data transfer to fit within the
allocated buffers.
Then use the sz command on Linux to send the file to the target:
$ sudo sz <filename> [-l nnnn] [-w nnnn] </dev/ttyS0 >/dev/ttyS0
Where <filename> is the file that you want to send. If -l nnnn and -w nnnn
is not specified, then there will likely be packet buffer overflow errors.
nnnn should be set to a value less than or equal to
CONFIG_SYSTEM_ZMODEM_PKTBUFSIZE
The resulting file will be found where you have configured the ZModem
"sandbox" via CONFIG_SYSTEM_ZMODEM_MOUNTPOINT.
You can add the sz -v option multiple times, each increases the level
of debug output. If you want to capture the Linux sz output, then
re-direct stderr to a log file by adding 2>sz.log to the end of the
sz command.
If you don't have the sz command on your Linux box, the package to
install rzsz (or possibily lrzsz).
Building the ZModem Tools to Run Under Linux
============================================
Build support has been added so that the NuttX ZModem implementation
can be executed on a Linux host PC. This can be done by
- Change to the apps/systems/zmodem directory
- Make using the special makefile, Makefile.host
NOTES:
1. TOPDIR and APPDIR must be defined on the make command line: TOPDIR is
the full path to the nuttx/ directory; APPDIR is the full path to the
apps/ directory. For example, if you installed nuttx at
/home/me/projects/nuttx and apps at /home/me/projects/apps, then the
correct make command line would be:
make -f Makefile.host TOPDIR=/home/me/projects/nuttx APPDIR=/home/me/projects/apps
2. Add CONFIG_DEBUG_FEATURES=1 to the make command line to enable debug output
3. Make sure to clean old target .o files before making new host .o files.
This build is has been verified as of 2013-7-16 using Linux to transfer
files with an Olimex LPC1766STK board. It works great and seems to solve
all of the problems found with the Linux sz/rz implementation.
Status
======
2013-7-15: Testing against the Linux rz/sz commands.
I have tested with the boards/arm/lpc17xx_40xx/olimex-lpc1766stk
configuration. I have been able to send large and small files with
the target sz command. I have been able to receive small files, but
there are problems receiving large files using the Linux sz command:
The Linux sz does not obey the buffering limits and continues to send
data while rz is writing the previously received data to the file and
the serial driver's RX buffer is overrun by a few bytes while the
write is in progress. As a result, when it reads the next buffer of
data, a few bytes may be missing. The symptom of this missing data is
a CRC check failure.
Either (1) we need a more courteous host application, or (2) we
need to greatly improve the target side buffering capability!
My thought now is to implement the NuttX sz and rz commands as
PC side applications as well. Matching both sides and obeying
the handshaking will solve the issues. Another option might be
to fix the serial driver hardware flow control somehow.
2013-7-16. More Testing against the Linux rz/sz commands.
I have verified that with debug off and at lower serial BAUD
(2400), the transfers of large files succeed without errors. I do
not consider this a "solution" to the problem. I also found that
the LPC17xx hardware flow control caused strange hangs; ZModem
works better with hardware flow control disabled on the LPC17xx.
At this lower BAUD, RX buffer sizes could probably be reduced; Or
perhaps the BAUD coud be increased. My thought, however, is that
tuning in such an unhealthy situation is not the approach: The
best thing to do would be to use the matching NuttX sz on the Linux
host side.
2013-7-16. More Testing against the NuttX rz/sz on Both Ends.
The NuttX sz/rz commands have been modified so that they can be
built and executed under Linux. In this case, there are no
transfer problems at all in either direction and with large or
small files. This configuration could probably run at much higher
serial speeds and with much smaller buffers (although that has not
been verified as of this writing).
2018-5-27:
Updates to checksum calculations. Verified correct operation with
hardware flow control using the olimex-stm32-p407/zmodem
configuration. Only the host-to-target transfer was verified.
This was using the Linux sz utility. There appears to still be a
problem using the NuttX sz utility running on Linux.
2018-5-27:
Verified correct operation with hardware flow control using the
olimex-stm32-p407/zmodem configuration with target-to-host
transfers was verified. Again, there are issues remaining if
I tried the NuttX rz utility running on Linux.
2018-6-26:
with -w nnnn option, the host-to-target transfer can work reliably
without hardware flow control.