254 lines
10 KiB
Plaintext
254 lines
10 KiB
Plaintext
SWO is a datastream that comes out of a single pin when the debug interface
|
|
is in SWD mode. It can be encoded either using NRZ (UART) or RZ (Manchester)
|
|
formats. The pin is a dedicated one that would be used for TDO when the
|
|
debug interface is in JTAG mode. On the STM32 it's port PB3.
|
|
|
|
When in NRZ mode the SWO data rate that comes out of the chip _must_ match
|
|
the rate that the debugger expects. By default on BMP the baudrate is
|
|
2.25MBps but that can be changed as an optional parameter to the monitor
|
|
traceswo command, like this;
|
|
|
|
monitor traceswo 115200
|
|
|
|
....would set the swo output at the low speed of 115kbps.
|
|
|
|
We are constrained on maximum input speed by both the capabilities of the
|
|
BMP STM32F103 USART and the ability to get the packets back out over the USB
|
|
link. The UART baudrate is set by b=(72x10^6)/(16*d)...so for d=1 that means
|
|
a maximum speed of 4.5Mbps. For continious streaming that turns out to be
|
|
_too_ fast for the USB link, so the next available option is the 2.25Mbps
|
|
that we use. ....you can safely use the 4.5Mbps setting if your debug data
|
|
is bursty, or if you're using a different CPU to the STM32F103 as your BMP
|
|
host, but you potentially run the risk of losing packets if you have long
|
|
runs of sending which the usb cannot flush in time (there's a 12K buffer, so
|
|
the it is a pretty long run before it becomes a problem).
|
|
|
|
Note that the baudrate equation means there are only certain speeds
|
|
available. The highest half dozen are;
|
|
|
|
1 4.50 Mbps
|
|
2 2.25 Mbps
|
|
3 1.50 Mbps
|
|
4 1.125 Mbps
|
|
5 0.900 Mbps
|
|
6 0.750 Mbps
|
|
|
|
...the USART will cope with some timing slip, but it's advisible to stay as
|
|
close to these values as you can. As the speed comes down the spread between
|
|
each valid value so mis-timing is less of an issue. The 'monitor traceswo
|
|
<x>' command will automatically find the closest divisor to the value you
|
|
set for the speed, so be aware the error could be significant.
|
|
|
|
Depending on what you're using to wake up SWO on the target side, you may
|
|
need code to get it into the correct mode and emitting data. You can do that
|
|
via gdb direct memory accesses, or from program code.
|
|
|
|
An example for a STM32F103 for the UART (NRZ) data format that we use;
|
|
|
|
/* STM32 specific configuration to enable the TRACESWO IO pin */
|
|
RCC->APB2ENR |= RCC_APB2ENR_AFIOEN;
|
|
AFIO->MAPR |= (2 << 24); // Disable JTAG to release TRACESWO
|
|
DBGMCU->CR |= DBGMCU_CR_TRACE_IOEN; // Enable IO trace pins
|
|
|
|
*((volatile unsigned *)(0xE0040010)) = 31; // Output bits at 72000000/(31+1)=2.25MHz.
|
|
*((volatile unsigned *)(0xE00400F0)) = 2; // Use Async mode (1 for RZ/Manchester)
|
|
*((volatile unsigned *)(0xE0040304)) = 0; // Disable formatter
|
|
|
|
/* Configure instrumentation trace macroblock */
|
|
ITM->LAR = 0xC5ACCE55;
|
|
ITM->TCR = 0x00010005;
|
|
ITM->TER = 0xFFFFFFFF; // Enable all stimulus ports
|
|
|
|
Code for the STM32L476 might look like:
|
|
#define BAUDRATE 115200
|
|
DBGMCU->CR |= DBGMCU_CR_TRACE_IOEN; /* Enable IO pins for Async trace */
|
|
uint32_t divisor, clk_frequency;
|
|
clk_frequency = NutGetCpuClock();
|
|
divisor = clk_frequency / BAUDRATE;
|
|
divisor--;
|
|
TPI->CSPSR = 1; /* port size = 1 bit */
|
|
TPI->ACPR = divisor;
|
|
TPI->SPPR = 2; /*Use Async mode pin protocol */
|
|
TPI->FFCR = 0x00; /* Bypass the TPIU formatter and send output directly*/
|
|
|
|
/* Configure Trace Port Interface Unit */
|
|
CoreDebug->DEMCR |= CoreDebug_DEMCR_TRCENA_Msk; // Enable access to registers
|
|
DWT->CTRL = 0x400003FE; // DWT needs to provide sync for ITM
|
|
ITM->LAR = 0xC5ACCE55; // Allow access to the Control Register
|
|
ITM->TPR = 0x0000000F; // Trace access privilege from user level code, please
|
|
ITM->TCR = 0x0001000D; // ITM_TCR_TraceBusID_Msk | ITM_TCR_DWTENA_Msk | ITM_TCR_SYNCENA_Msk | ITM_TCR_ITMENA_Msk
|
|
ITM->TER = 1; // Only Enable stimulus port 1
|
|
|
|
while(1) {
|
|
for (uint32_t i = 'A'; i <= 'Z'; i++) {
|
|
ITM_SendChar(i);
|
|
NutSleep(1);
|
|
}
|
|
}
|
|
|
|
If you're using RZ mode (e.g. on a genuine BMP) then you will need the trace
|
|
output speed to be quite a lot lower...in the order of 200kHz by means of
|
|
changing the divisor to something like 359. That's because the STM32F103
|
|
doesn't have a dedicated RZ decoder so it all has to be done in
|
|
software. The advantage of RZ is that the probe can adapt to the speed of
|
|
the target, so you don't have to set the speed on the probe in the monitor
|
|
traceswo command, and it will be tolerant of different speeds.
|
|
|
|
The SWO data appears on USB Interface 5, Endpoint 5.
|
|
|
|
SWOListen
|
|
=========
|
|
A program swolisten.c is found in ./scripts which will listen to this
|
|
endpoint, decode the datastream, and output it to a set of unix fifos which
|
|
can then be used as the input to other programs (e.g. cat, or something more
|
|
sophisticated like gnuplot, octave or whatever). This program doesn't care
|
|
if the data originates from a RZ or NRZ port, or at what speed.
|
|
|
|
Note that swolisten can be used with either BMP firmware, or with a
|
|
conventional TTL serial dongle. See at the bottom of this file for
|
|
information on how to use a dongle.
|
|
|
|
The command line to build the swolisten tool is;
|
|
|
|
gcc -I /usr/local/include/libusb-1.0 -L /usr/local/lib -lusb-1.0 swolisten.c -o swolisten
|
|
|
|
For Opensuse:
|
|
gcc -I /usr/include/libusb-1.0 -lusb-1.0 swolisten.c swolisten -std=gnu99 -g -Og
|
|
|
|
...you will obviously need to change the paths to your libusb files.
|
|
|
|
Attach to BMP to your PC:
|
|
Start gdb: "arm-none-eabi-gdb"
|
|
Choose bmp as target, like:
|
|
"target extended /dev/ttyACM0(*)"
|
|
Start SWO output: "mon traceswo"
|
|
If async SWO is used, give the baudrate your device sends
|
|
out as argument. 2.25 MBaud is the default, for the STM32L476 example above
|
|
the command would be: "mon traceswo 115200(*)".
|
|
Scan the SWD "mon swdp_scan"
|
|
Attach to the device: : "attach 1"
|
|
Start the program: "r".
|
|
(*) Your milage may vary
|
|
Now start swolisten without further options.
|
|
|
|
By default the tool will create fifos for the first 32 channels in a
|
|
directory swo (which you will need to create) as follows;
|
|
|
|
>ls swo/
|
|
chan00 chan02 chan04 chan06 chan08 chan0A chan0C chan0E chan10 chan12 chan14
|
|
chan16 chan18 chan1A chan1C chan1E chan01 chan03 chan05 chan07 chan09 chan0B
|
|
chan0D chan0F chan11 chan13 chan15 chan17 chan19 chan1B chan1D chan1F
|
|
|
|
>cat swo/channel0
|
|
<<OUTPUT FROM ITM Channel 0>>
|
|
|
|
With the F103 and L476 examples above, an endless stream of
|
|
"ABCDEFGHIJKLMNOPQRSTUVWXYZ" should be seen. During reset of the target
|
|
device, no output will appear, but with release of reset output restarts.
|
|
|
|
Information about command line options can be found with the -h option.
|
|
swolisten is specifically designed to be 'hardy' to probe and target
|
|
disconnects and restarts (y'know, like you get in the real world). The
|
|
intention being to give you streams whenever it can get them. It does _not_
|
|
require gdb to be running. For the time being traceswo is not turned on by
|
|
default in the BMP to avoid possible interactions and making the overall
|
|
thing less reliable so You do need gdb to send the initial 'monitor
|
|
traceswo' to the probe, but beyond that there's no requirement for gdb to be
|
|
present.
|
|
|
|
Reliability
|
|
===========
|
|
|
|
A whole chunk of work has gone into making sure the dataflow over the SWO
|
|
link is reliable. The TL;DR is that the link _is_ reliable. There are
|
|
factors outside of our control (i.e. the USB bus you connect to) that could
|
|
potentially break the reliabilty but there's not too much we can do about
|
|
that since the SWO link is unidirectional (no opportunity for
|
|
re-transmits). The following section provides evidence for the claim that
|
|
the link is good;
|
|
|
|
A test 'mule' sends data flat out to the link at the maximum data rate of
|
|
2.25Mbps using a loop like the one below;
|
|
|
|
while (1)
|
|
{
|
|
for (uint32_t r=0; r<26; r++)
|
|
{
|
|
for (uint32_t g=0; g<31; g++)
|
|
{
|
|
ITM_SendChar('A'+r);
|
|
}
|
|
ITM_SendChar('\n');
|
|
}
|
|
}
|
|
|
|
100MB of data (more than 200MB of actual SWO packets, due to the encoding) was sent from the mule to the BMP where the
|
|
output from swolisten chan00 was cat'ted into a file;
|
|
|
|
>cat swo/chan00 > o
|
|
|
|
....this process was interrupted once the file had grown to 100MB. The first
|
|
and last lines were removed from it (these represent previously buffered
|
|
data and an incomplete packet at the point where the capture was
|
|
interrupted) and the resulting file analysed for consistency;
|
|
|
|
> sort o | uniq -c
|
|
|
|
The output was;
|
|
|
|
126462 AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
126462 BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
|
|
126462 CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC
|
|
126462 DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD
|
|
126461 EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE
|
|
126461 FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
|
|
126461 GGGGGGGGGGGGGGGGGGGGGGGGGGGGGGG
|
|
126461 HHHHHHHHHHHHHHHHHHHHHHHHHHHHHHH
|
|
126461 IIIIIIIIIIIIIIIIIIIIIIIIIIIIIII
|
|
126461 JJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ
|
|
126461 KKKKKKKKKKKKKKKKKKKKKKKKKKKKKKK
|
|
126461 LLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL
|
|
126461 MMMMMMMMMMMMMMMMMMMMMMMMMMMMMMM
|
|
126461 NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN
|
|
126461 OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
|
|
126461 PPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP
|
|
126461 QQQQQQQQQQQQQQQQQQQQQQQQQQQQQQQ
|
|
126461 RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRR
|
|
126461 SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS
|
|
126461 TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT
|
|
126461 UUUUUUUUUUUUUUUUUUUUUUUUUUUUUUU
|
|
126461 VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV
|
|
126461 WWWWWWWWWWWWWWWWWWWWWWWWWWWWWWW
|
|
126461 XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
|
|
126461 YYYYYYYYYYYYYYYYYYYYYYYYYYYYYYY
|
|
126461 ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
|
|
|
|
(On inspection, the last line of recorded data was indeed a 'D' line).
|
|
|
|
Swolisten, using a TTL Serial Dongle
|
|
====================================
|
|
|
|
The NRZ data that comes out of the SWO is just UART formatted, but in a
|
|
frame. swolisten has been extended to accomodate TTL Serial Dongles that
|
|
can pick this up. Success has been had with CP2102 dongles at up to 921600
|
|
baud.
|
|
|
|
To use this mode just connect SWO to the RX pin of your dongle, and start
|
|
swolisten with parmeters representing the speed and port. An example;
|
|
|
|
>./swolisten -p /dev/cu.SLAB_USBtoUART -v -b swo/ -s 921600
|
|
|
|
Any individual dongle will only support certain baudrates (Generally
|
|
multiples of 115200) so you may have to experiment to find the best
|
|
supported ones. For the CP2102 dongle 1.3824Mbps wasn't supported and
|
|
1.8432Mbps returned corrupted data.
|
|
|
|
Please email dave@marples.net with information about dongles you find work
|
|
well and at what speed.
|
|
|
|
Further information
|
|
===================
|
|
SWO is a wide field. Read e.g. the blogs around SWD on
|
|
http://shadetail.com/blog/swo-starting-the-steroids/
|
|
An open source program suite for SWO under active development is
|
|
https://github.com/mubes/orbuculum |