Jetson Nano (Developer Kit) SPI Hell

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

Hi,

So I’ve read all I could read.
Went through the instructions on https://github.com/rt-net/JetsonNano_DT_SPI
The script executed successfully from a host running Ubuntu. After the flashing was over and the Jetson restarted I got ‘spidev’ when I did ‘lsmod’.
I was also able to see spidev0.0 and spidev1.0 under /dev.

At this point I took the kernel test code spidev.c compiled it and executed '/spidev_test -D /dev/spidev0.0 -v
Placed a probe on PIN 23 (CLK) and …

Nothing :)
Absolutely nothing at all.
NVIDIA, letting developers an easy access to peripherals (SPI, I2C. UART act’) is mandatory, in my opinion, for the success of this line of products.

Eitan.

2

If you’re running JetPack 4.3, Linux For Tegra 32.3.1, use the new jetson-io tool to enable spidev.

https://docs.nvidia.com/jetson/archives/l4t-archived/l4t-3231/index.html#page/Tegra%2520Linux%2520Driver%2520Package%2520Development%2520Guide%2Fhw_setup_jetson_io.html%23

If running an older version, try the following…

https://devtalk.nvidia.com/default/topic/1062646/jetson-nano/updated-instructions-for-spi-on-nano-developmentkit-with-l4t-32-2-1/1

The instructions change during that thread so read all the way through.

3

Since I have a clean image that I’ve DL today I’ve tested the python tool.
As expected I got the error, I’ve typed the magical line :

sudo find /opt/nvidia/jetson-io/ -mindepth 1 -maxdepth 1 -type d -exec touch {}/init.py ;

And tested again the python script (sudo /opt/nvidia/jetson-io/jetson-io.py ).
I was able to see something that blinks momentarily and then it was gone.
No error message, no nothing.

Any clue ?

4

You also need to make sure the existing DTBs are both in the /boot and /boot/dtb directories.

You can see the error by running

jetson-io.py | less

This removes the screen formatting so you can see the error at the bottom of the output.

5

Thanks,

And how shall I get those files?
Ans yes, I known It’s probably buried deep in one of the gazillion readme files in the SDK.

6

They’re already in /boot. You just need to create a /boot/dtb directory and copy them down.

7

Thanks !!

SPI (master) works, the simple spidev.c tester works and I can see the clocks from an external logic.
Now, I’m trying to make the Jetson nano work in a slave mode, so it could read and analyze a fast stream of data being sent to it from an external sensor,

So the next question would be, did this python script enabled SPI slave on SPI 0 & 1?
If so, could you point me to a simple SPI slave source that I could use as a reference, My focus is more on analyzing the data in the Jetson core rather the communication method which seem to take most of my time.

Thanks for the help!
Eitan.

8

There currently isn’t a device tree model for using the tegra210 in slave mode. There’s a driver that supports it though so I can try and create one. Give me a day or so.

9

That would be awesome,
Thank you very much!

10

I think I’ve got a devicetree overlay that’ll put a controller in slave mode. I’m testing and packaging now. Do you have some software that uses the slave mode that I could test with?

11

Well, this is exactly what I have to create.

The master in this case sends messages of 256 / 512 bytes, where the slave (Jetson) never sends anything, it’s a simple SPI listener, not sure if the CS is being used at all.
Each ‘message’ has start byte of 0xa5 and the last 2 bytes are crc16.
Once I have SPI slave mode running all I’ll crate some sort of a Q that stores those messages and another thread that takes those messages and do all the math.

12

Hi,
Any chance there’s something new with that slave overlay?

13

Sorry. I got carried away with the install procedure and wound up writing a whole python library to create overlays. :) I’ll have a zip file for you to test with later this afternoon (GMT-7).

14

OK, here you go…

# unzip DTBOverlayCreator.zip
# cd DTBOverlayCreator
# make
# make install
# reboot

I’m still working on the documentation but running SPI_Custom.py will create a dts file that enables /dev/spidev0.0 in master mode and /dev/spidev1.0 in slave mode. The Makefile runs that python script and compiles the result into a new dtb file.

Give it a shot and see how it works.
DTBOverlayCreator.zip (25 KB)

15

Thanks!

OK,so I have my sensor hooked to SPI1 (hope I got it right)
SPI1_CLK : PIN 13
SPI1_MISO : PIN 22
SPI1 CS0 : PIN 18
SPI1 MOSI : None, we are the slave and we never talk.

I have executed your script and rebooted.
So, since the overlay enables SPI on spidev I could actually read data from the master in user space ?
( I can see only spi=master under /sys/class)

If so could you point me to a sample that does this?
Could find anything regarding Linux, SPI -Slave, SPIDev.

Thanks
Eitan.

16

the spi device will still show as a “master” device even though it’s operating in slave mode.

This should read 1 byte from /dev/spidev1.0 and print it. Set the block_size, mode, bits, and speed as appropriate.

#include <stdint.h>
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <linux/ioctl.h>
#include <sys/stat.h>
#include <linux/types.h>
#include <linux/spi/spidev.h>

int main() {
	int block_size = 1;
	uint32_t mode = 0;
	uint8_t bits = 8;
	uint32_t speed = 1000000;
	uint8_t rx[block_size];
	int ret;
	int fd;


	fd = open("/dev/spidev1.0", O_RDONLY);

	// Example mode. Just use the modes you need if any.
	// mode = SPI_CPHA | SPI_CPOL | SPI_LSB_FIRST | SPI_LSB_FIRST | SPI_NO_CS;
	ret = ioctl(fd, SPI_IOC_WR_MODE32, &mode);
	ret = ioctl(fd, SPI_IOC_WR_BITS_PER_WORD, &bits);
	ret = ioctl(fd, SPI_IOC_WR_MAX_SPEED_HZ, &speed);

	ret = read(fd, rx, block_size);
	
	printf("0x%2x\n", rx[0]);

	return 0;

}
17

Hello,

First, I can’t thank you enough for all the help !
So I have the following :

I’ve attached a different board to the Jetosn’s SPI-1 (only Clk, MOSI & CS). This board is a
master that sends 0xa5 and 0x5a continuously at 500Khz, I have a logic that observing the data being sent between that board and the Jetson, everything looks fine, CS goes low, master is sending data during the 8 clocks it’s generating.

I’ve used SPI1 (37: MOSI, 13 CLK, 18: CS).
I’ve used you reference code, and made an infinite loop that read and print a received byte.
In another terminal I have ‘dmesg -w’
Immediately after I turn the master device on I get :

[ 8283.765308] spi_master spi1: CMD[03f01027]: Sl M0 CS0 [HHHH] MSB MSb Rx Pa 8b TRANS[00ff0000]:BSY I:255 B:0
FIFO[00c00004]:RxF:1 TxE:64 Err RxSTA TxSTA[E]DMA[00000000]: RxTr:0 TxTr:0 B:0
[ 8283.784388] spi_master spi1: failed to transfer one message from queue
[ 8283.792313] spi_master spi1: failed to transfer one message from queue
[ 8283.800204] spi_master spi1: failed to transfer one message from queue
[ 8283.807625] spi_master spi1: failed to transfer one message from queue … endlessly

(Doesn’t look good but it means that the Jetson knows that data is being transmitted, when I turn the master device off, those error messages stops).

My process gets many errors (read returns a negative value) and once in a while it reads a byte (but not 0xa5 or 0xa5).

Any clue?
Eitan.

18

Forgot to ad:
I’ve tested this at a very low speed (1 bytes every ~2 seconds at ~100Khz).
The full (repeating) error sounds like a DMA issue :

[12712.352737] spi_master spi1: rx-dma-err [status:80c00004]
[12712.358568] spi_master spi1: CMD[03f01027]: Sl M0 CS0 [HHHH] MSB MSb Rx Pa 8b TRANS[00ff0001]:BSY I:255 B:1
FIFO[00c00004]:RxF:1 TxE:64 Err RxSTA TxSTA[E]DMA[00110000]: RxTr:2 TxTr:2 B:4095
[12712.378533] spi_master spi1: failed to transfer one message from queue

Thanks,
Eitan.

19

Well, that’s progress. :)

Try adding SPI_CPHA to the mode.

20

Something is defiantly working now :)
And I think I know what’s causing the ‘failed to transfer one message from queue’ errors
My device is sending data at 12Mhz with about 30μ between each byte.
if I place ~5ms gap between the bytes everything works, once the gap is less I get tons of errors.

Is there anything I could do ?
The real sensor is sending data at fast rate and I can’t change it’s behavior.