On Tue, Dec 17, 2024 at 09:08:23PM +0100, Lode Willems wrote: > On 14/12/2024 18:21, Johan Hovold wrote: > > On Wed, Nov 13, 2024 at 07:08:27PM +0100, Lode Willems wrote: > >> This adds support for enabling and disabling > >> RTS/CTS hardware flow control. > >> Tested using a CH340E in combination with a CP2102. > >> > >> Fixes part of the following bug report: > >> Link: https://bugzilla.kernel.org/show_bug.cgi?id=197109 > >> > >> Signed-off-by: Lode Willems <me@xxxxxxxxxxxxxxx> > > > > Thanks for the patch and sorry about the late feedback on this one. I > > wanted to give it a spin with the devices I have here (ch340g and > > ch341a). > > Thanks for the review. Sorry in advance if this e-mail is formatted > incorrectly, this is my first time replaying to a review. Looks all good. > > Appears to the modem control lines are not wired up on the ch341a > > evaluation board I have, but the device accepts the request and stops > > transmitting with hardware flow control enabled. > > Since creating this patch I've acquired a ch341a breakout board and can > confirm that it transmits with hardware flow control enabled and CTS > asserted. Thanks for checking. > > With ch340g in loopback, I also see it refuse to transmit unless cts is > > asserted. I was not able to get the device to deassert rts when > > attempting to fill its receive buffers, however. Perhaps the hardware > > does not support this, but is this something you tried? > > I didn't try this before. Just trying a couple of things now I also > couldn't seem to make it deassert RTS, but this may be a failure of > how I'm testing this. Or it's a limitation of the hardware (e.g. it really only support auto-cts). I patched the driver to not submit any read urbs, which should eventually fill up the receive buffers. > >> +#define CH341_FLOW_CTL_NONE 0x000 > >> +#define CH341_FLOW_CTL_RTSCTS 0x101 > > > > The registers are eight bits wide, but the driver writes two at a time. > > So this should just be 0x00 and 0x01. > > Ok. I'm guessing I don't have to send the same request twice and can > just leave the top eight bits of the value and index empty? > It seems to work in the quick testing I've done. Possibly, but writing the same register twice is what the vendor driver is doing here (e.g. otherwise you may actually write to offset 0 as well). > >> +static void ch341_set_flow_control(struct tty_struct *tty, > >> + struct usb_serial_port *port, > >> + const struct ktermios *old_termios) > >> +{ > >> + int r; > >> + > >> + if (old_termios && > >> + C_CRTSCTS(tty) == (old_termios->c_cflag & CRTSCTS)) > >> + return; > > > > Just drop this and set the requested setting unconditionally. > > Ok > > > > >> + > >> + if (C_CRTSCTS(tty)) { > >> + r = ch341_control_out(port->serial->dev, CH341_REQ_WRITE_REG, > >> + CH341_REG_FLOW_CTL | > >> + ((u16)CH341_REG_FLOW_CTL << 8), > >> + CH341_FLOW_CTL_RTSCTS); > >> + if (r < 0) { > >> + dev_err(&port->dev, > >> + "%s - failed to enable flow control: %d\n", > >> + __func__, r); > >> + tty->termios.c_cflag &= ~CRTSCTS; > >> + } > >> + } else { > >> + r = ch341_control_out(port->serial->dev, CH341_REQ_WRITE_REG, > >> + CH341_REG_FLOW_CTL | > >> + ((u16)CH341_REG_FLOW_CTL << 8), > >> + CH341_FLOW_CTL_NONE); > >> + if (r < 0) { > >> + dev_err(&port->dev, > >> + "%s - failed to disable flow control: %d\n", > >> + __func__, r); > >> + tty->termios.c_cflag |= CRTSCTS; > >> + } > >> + } > > > > Please rewrite this so that you prepare the value and index parameters > > based on the termios settings and then do one control request. If it > > fails you can restore the old setting (if old_termios is non-NULL). > > Ok, I haven't seen any other driver restore the old setting on request > failure, now I'm questioning if it's actually necessary? Not all drivers do, but it is the right thing to do as that's the only way we have to report failure to change a setting. > If it is, I'll change it to the following: > tty->termios.c_cflag = (tty->termios.c_cflag & !CRTSCTS) You meant bitwise negation here (~) as you used above. > | (old_termios->c_cflag & CRTSCTS); But remember to only do this if old_termios is non-NULL. Might be cleaner as two separate statements (&= and |=). > > And please drop the redundant __func__ from the error message (even if > > the driver still uses it in some other functions still). > > Ok. Looking at the code again, the error already gets logged in > ch341_control_out on failure. Maybe this log line shouldn't be added? Yeah, I guess that's more in line with what the driver is currently doing. We could always amend those printk to include index and value to make it easier to figure out which call is failing if this becomes an issue. Johan
