Jetson Orin Nano PWM only works on Pin 33

[mattwilliamson]

Hello,

Pin 33 works with PWM. Pin 32 and pin 15 do not. It's just either on or off. I'm using a L293 motor controller and if I switch the pins for left and right motors, the other motor exhibits the same behavior.

Jetson Orin Nano
Jetpack 5.1.2 L4T 35.4.1
Jetson.GPIO.VERSION '2.1.6'

>>> GPIO.JETSON_INFO
{'P1_REVISION': 1, 'RAM': '32768M, 65536M', 'REVISION': 'Unknown', 'TYPE': 'JETSON_ORIN_NANO', 'MANUFACTURER': 'NVIDIA', 'PROCESSOR': 'A78AE'}

Here's my sample code:

import Jetson.GPIO as GPIO
import time

GPIO.setwarnings(False)
GPIO.setmode(GPIO.BOARD)

# Enable, forward, backward
lpin = [32, 38, 37]
rpin = [33, 35, 36]
hz = 50

GPIO.setup(lpin[0], GPIO.OUT)
GPIO.setup(rpin[0], GPIO.OUT)
GPIO.setup(lpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(lpin[2], GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[2], GPIO.OUT, initial=GPIO.LOW)

lpwm = GPIO.PWM(lpin[0], hz)
rpwm = GPIO.PWM(rpin[0], hz)

lpwm.start(0)
rpwm.start(0)

#

def forward(left_speed=100, right_speed=100):
    lpwm.ChangeDutyCycle(left_speed)
    rpwm.ChangeDutyCycle(right_speed)
    GPIO.output(lpin[1],GPIO.HIGH)
    GPIO.output(rpin[1],GPIO.HIGH)
    GPIO.output(lpin[2],GPIO.LOW)
    GPIO.output(rpin[2],GPIO.LOW)

for i in range(0, 100, 1):
    print(i)
    forward(i, 0)
    time.sleep(.1)

Some more info for you:

At 50% duty cycle on pin 32 and 33:

matt@deepdrive:~$ sudo cat /sys/kernel/debug/pwm
platform/39c0000.tachometer, 1 PWM device
 pwm-0   ((null)              ): period: 0 ns duty: 0 ns polarity: normal

platform/32e0000.pwm, 1 PWM device
 pwm-0   (sysfs               ): requested enabled period: 20000000 ns duty: 0 ns polarity: normal

platform/32c0000.pwm, 1 PWM device
 pwm-0   (sysfs               ): requested enabled period: 20000000 ns duty: 10000000 ns polarity: normal

platform/32a0000.pwm, 1 PWM device
 pwm-0   (pwm-fan             ): requested enabled period: 45334 ns duty: 14045 ns polarity: normal

platform/3280000.pwm, 1 PWM device
 pwm-0   (sysfs               ): requested enabled period: 20000000 ns duty: 0 ns polarity: normal

[mattwilliamson]

Once python is running with GPIO, this seems to be able to control the duty cycle in another shell:

echo 0 > /sys/class/pwm/pwmchip3/export
echo 20000000 > /sys/class/pwm/pwmchip3/pwm0/period
echo 10000000 > /sys/class/pwm/pwmchip3/pwm0/duty_cycle
echo 1 > /sys/class/pwm/pwmchip3/pwm0/enable

[anhmiuhv]

Did you change the pinmux so they are pwm?

[mattwilliamson]

Yep. I don't think the /sys/class/pwm/pwmchip3/pwm0/duty_cycle would have worked without that.

[anhmiuhv]

Does the code run without crashing and the output signal is correct for pin 33 but the output signal is not correct for pin 15 and 32?

[anhmiuhv]

Once python is running with GPIO, this seems to be able to control the duty cycle in another shell:

echo 0 > /sys/class/pwm/pwmchip3/export
echo 20000000 > /sys/class/pwm/pwmchip3/pwm0/period
echo 10000000 > /sys/class/pwm/pwmchip3/pwm0/duty_cycle
echo 1 > /sys/class/pwm/pwmchip3/pwm0/enable

Yes actually this is how the underlying python code control GPIO PWM. You can do this to control it manually

[mattwilliamson]

Does the code run without crashing and the output signal is correct for pin 33 but the output signal is not correct for pin 15 and 32?

Yes, that's correct.

[mattwilliamson]

There's no software PWM support, right? I'm just thinking of workarounds until this can get fixed.

[anhmiuhv]

No software PWM support. Maybe you can use external servo driver like this https://www.amazon.com/16-Channel-12-bit-Servo-Driver-Interface/dp/B00EIB0U7A

[anhmiuhv]

Can you show me the output of

sudo busybox devmem 0x02440020 sudo busybox devmem 0x02434080

[mattwilliamson]

$ sudo busybox devmem 0x02440020
0x00000404
$ sudo busybox devmem 0x02434080
0x00000404

[anhmiuhv]

I cannot reproduce this on my hardware. I will let our hardware engineer know to look into this further.

[mattwilliamson]

If you have instructions on how you got yours working, I could double check that I'm doing it correctly.

[anhmiuhv]

I tested PWM using https://github.com/NVIDIA/jetson-gpio/blob/master/samples/test_all_apis.py To run it you need to connect pin 33 - 19 and pin 11 - 13, and set up pin 15,32,33 as PWM . I modify the test https://github.com/NVIDIA/jetson-gpio/blob/af27a7db4adca91f40fb1aac75a730e7f9c5ba4a/samples/test_all_apis.py#L38 so that "33" is replaced with 15 and 32, and reconnect the jumper wire correspondingly to see if PWM work correctly for those pin

[mattwilliamson]

Nice! I will try it in the morning.

[mattwilliamson]

Here is one thing to note:

>>> GPIO.cleanup()
OSError: [Errno 9] Bad file descriptor

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/matt/src/deepdrive/venv/lib/python3.8/site-packages/Jetson/GPIO/gpio.py", line 381, in cleanup
    _cleanup_all()
  File "/home/matt/src/deepdrive/venv/lib/python3.8/site-packages/Jetson/GPIO/gpio.py", line 273, in _cleanup_all
    _cleanup_one(ch_info)
  File "/home/matt/src/deepdrive/venv/lib/python3.8/site-packages/Jetson/GPIO/gpio.py", line 248, in _cleanup_one
    _unexport_pwm(ch_info)
  File "/home/matt/src/deepdrive/venv/lib/python3.8/site-packages/Jetson/GPIO/gpio.py", line 201, in _unexport_pwm
    ch_info.f_duty_cycle.close()
OSError: [Errno 9] Bad file descriptor

[mattwilliamson]

test_all_apis.py passed with pin32.

Am I correct in that to test pin 15, I connect pin 15 to 19 instead of 33 - 19?

If I do that, I see this:

Testing test_warnings_off
Testing test_warnings_on
Testing test_setup_one_board
Testing test_setup_one_bcm
Testing test_setup_one_cvm
Testing test_setup_one_tegra_soc
Testing test_setup_twice
Testing test_setup_one_out_no_init
Testing test_setup_one_out_high
Testing test_setup_one_out_low
Testing test_setup_many_out_no_init
Testing test_setup_many_out_one_init
Testing test_setup_many_out_many_init
Testing test_setup_one_in
Testing test_setup_one_in_pull
/home/matt/src/deepdrive/venv/lib/python3.8/site-packages/Jetson/GPIO/gpio.py:345: UserWarning: Jetson.GPIO ignores setup()'s pull_up_down parameter
  warnings.warn("Jetson.GPIO ignores setup()'s pull_up_down parameter")
Testing test_setup_many_in
Testing test_setup_all
Testing test_cleanup_one
Testing test_cleanup_many
Testing test_cleanup_all
Testing test_input
Testing test_output_one
Testing test_output_many_one_value
Testing test_output_many_many_value
Testing test_out_in_init_high
Testing test_out_in_init_low
Testing test_gpio_function_unexported
Testing test_gpio_function_in
Testing test_gpio_function_out
Testing test_wait_for_edge_timeout
Testing test_wait_for_edge_rising
Testing test_wait_for_edge_falling
Testing test_event_detected_falling
Testing test_event_detected_rising
Testing test_event_detected_both
Testing test_event_callbacks
Testing test_multi_events_detected_diff_edge
Testing test_multi_events_detected_same_edge
Testing test_pwm_multi_duty
Traceback (most recent call last):
  File "test_all_apis.py", line 1084, in <module>
    test()
  File "test_all_apis.py", line 1009, in test_pwm_multi_duty
    assert min_ct <= count <= max_ct
AssertionError

[mattwilliamson]

Oh it seems that I figured it out. It's quite interesting. Order matters. If I do all the pin setup one at a time it is working.

This will NOT WORK:

GPIO.setup(lpin[0], GPIO.OUT)
GPIO.setup(rpin[0], GPIO.OUT)

GPIO.setup(lpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(lpin[2], GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[2], GPIO.OUT, initial=GPIO.LOW)

lpwm = GPIO.PWM(lpin[0], hz)
rpwm = GPIO.PWM(rpin[0], hz)

lpwm.start(0)
rpwm.start(0)

This WILL work:

GPIO.setup(lpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[1],GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(lpin[2], GPIO.OUT, initial=GPIO.LOW)
GPIO.setup(rpin[2], GPIO.OUT, initial=GPIO.LOW)

# Setup this pin and pwm first
GPIO.setup(lpin[0], GPIO.OUT)
lpwm = GPIO.PWM(lpin[0], hz)
lpwm.start(0)

# then this one
GPIO.setup(rpin[0], GPIO.OUT)
rpwm = GPIO.PWM(rpin[0], hz)
rpwm.start(0)

I'm not sure if it's a bug or if it's a hardware limitation that should be documented.