Re: [PATCH] usb-serial: wakeup device on input

[Jason Andryuk]

On Mon, Mar 9, 2020 at 6:08 AM Gerd Hoffmann <address@hidden> wrote:
>
> On Fri, Mar 06, 2020 at 09:09:17AM -0500, Jason Andryuk wrote:
> > Currently usb-serial devices are unable to send data into guests with
> > the xhci controller.  Data is copied into the usb-serial's buffer, but
> > it is not sent into the guest.  Data coming out of the guest works
> > properly.  usb-serial devices work properly with ehci.
> >
> > Have usb-serial call usb_wakeup() when receiving data from the chardev.
> > This seems to notify the xhci controller and fix inbound data flow.
> >
> > Also add USB_CFG_ATT_WAKEUP to the device's bmAttributes.  This matches
> > a real FTDI serial adapter's bmAttributes.
> >
> > Signed-off-by: Jason Andryuk <address@hidden>
>
> Added to usb patch queue.

Thanks.  Unfortunately, while this seemed okay in my early testing,
something is still off.  Typing at slow (human) speed, input seems
fine.  Pasting a large chunk of data into netcat has some of the data
dropped (not coming out of /dev/ttyUSB0 in the guest).  The VM kernel
reports "ttyUSB0: X input overrun(s)" with X seen between 1 and 8.
EHCI seems fine, even for chunks greater than the 384 byte buffer of
usb-serial.

As one example, pasting
Aa0Aa1Aa2Aa3Aa4Aa5Aa6Aa7Aa8Aa9Ab0Ab1Ab2Ab3Ab4Ab5Ab6Ab7Ab8Ab9Ac0Ac1Ac2Ac3Ac4Ac5Ac6Ac7Ac8Ac9Ad0Ad1Ad2Ad3Ad4Ad5Ad6Ad7Ad8Ad9Ae0
looks to only have the following output in the guest
Aa0Aa1Aa2Aa3Aa4Aa5Aa6Aa7Aa8Aa9Ab0Ab1Ab2Ab3Ab4Ab5Ab6Ab7Ab8Ab9Acc1Ac2Ac3Ac4Ac5Ac6Ac7Ac8Ac9Ad0Ad1Ad2Ad3Ad4Ad5Ad6Ad7Ad8Ad9Ae0

Characters 62 & 63 don't make it through "c0Ac" -> "cc".  The guest
kernel does *not* report input overruns in this case.

Any ideas?

Thanks,
Jason