I implemented USB Device feature for Renesas mbed boards.
The code referenced the following code as a starting point and is implemented it by inheritting USBPhy same as other boards.
(mbed-os\features\unsupported\USBDevice\targets\TARGET_RENESAS)
A DTR line is used to signal that the host has configured a terminal and
is ready to transmit and receive data from the USB CDC/Serial device.
When this test suite is run with the use of a Linux host, a workaround has
to be used to overcome some platform specific DTR line behavior.
Every time the serial port file descriptor is opened, the DTR line is
asserted until the terminal attributes are set.
As a consequence, the device receives a premature DTR signal with a
duration of 200-500 us before the correct, long-lasting DTR signal set by
the host-side test script. (tested on the Linux kernel 4.15.0)
The solution is to wait for the first DTR spike, ignore it, and wait for
the correct DTR signal again.
Update the serial host test to set DTR low before opening a serial
port. Set DTR high once the serial object is ready. This ensures that
no data is sent by the device until the host is ready.
Despite having ECHO and ECHOCTL POSIX lflags disabled by default by
pyserial, a delay is needed for host to properly handle data received
from USB serial/CDC device.
With no delay host would echo 0x00-0x31 characters prefixed with '^'.
Ignore interrupts on disabled USB endpoints. This prevents handling
interrupts when in the wrong state.
Prior to this patch when running the serial test on a K64F the assert
on line 908 of USBDevice.cpp would sometimes be triggered. This
assert indicates that an endpoint 0 IN interrupt occurred before
the device was ready.
This occurs during the test_cdc_usb_reconnect test when the host sends
a "Set Control Line State" USB request and the device acknowledges it
just before USB is disconnected.
Although the USB spec sets the upper limit on FS isochronous endpoint
payloads to 1023 B, this value is hard to test in practice. Moreover,
not all the targets Mbed OS supports (like NUCLEO_F207ZG) are able to
handle all the endpoints set to max.
Fix the host script issues present on Windows machines.
Add 0 B payload size to bulk endpoints test.
Update halt and abort tests according to PR comments.
Add an explicit request to start reading on OUT endpoints.
When activating an endpoint assign new data rather than ORing
data to it. This ensures that values set from the previous use
do not effect the current configuration.
Russ Butler
bcostm
Filip Jagodzinski
TomoYamanaka
Filip Jagodzinski