On Tue, Jun 09, 2020 at 01:13:06PM +0000, Joakim Tjernlund wrote: > On Tue, 2020-06-09 at 13:57 +0200, Greg KH wrote: > > > > On Tue, Jun 09, 2020 at 11:38:49AM +0000, Joakim Tjernlund wrote: > > > Hi List > > > > > > I was advised to come here with this problem(started on the USB list). > > > > > > We have a USB to RS232 bridge which presents itself as an ttyACM and the first connect after power on, > > > we see some garbage chars transmitted back from USB host(PC) to out device which becomes input to > > > the device. > > > > > > After much debugging I found that this are chars sent early in the boot process which then > > > are buffered and the TTYs default to ECHO chars is the cause. > > > > So some program in the boot sequence is trying to send data out the > > device? Why not just not do that? > > This is the boot console. Both u-boot and Linux prints a lot there, then init prints while starting services So the same device is used for boot console as well as a ttyACM device later on? > > > When the TTY is opened, any chars in the this buffer is ECHOed back over USB to the device, > > > before one has a chance to disable ECHO. The device then thinks these chars are regular input. > > > > Wait, you said something in the boot process did write to the device, > > which would have caused the tty to be opened then, right? > > well, boot process of the device prints and it is enough for the USB cable to be attached but not opened by any app. > The device just see an UART and prints when UART is initialized. What tool does that? Why not fix that? > > > Seems to me that this behaviour is unwanted in general and and app. should get a chance to flush/discard > > > any chars so this does not happen. > > > > Where are the characters coming from that would need to be flushed? > > Early output from boot, basically whatever prints just after connecting the USB cable. Then don't have boot print to that device :) > > When should characters be flushed exactly? > > Whatever is in the buffers before opening the tty. But what is supposed to happen to the data that was sent to it while it was "closed"? > The terminal app(like cu) tries to flush any input when it starts, just to avoid any old chars in the > queue but it is to late then. I strongly just suggest having userspace not write to the device to start with, that would solve this, right? thanks, greg k-h