Re: 2.6.31-rc5 regression: Oops when USB Serial disconnected while in use

[Greg KH <gregkh@suse.de>]

On Mon, Aug 10, 2009 at 08:51:15PM +0200, Bruno Prémont wrote:
> On Mon, 10 August 2009 Bruno Prémont <bonbons@linux-vserver.org> wrote:
> > On Mon, 10 August 2009 Greg KH <greg@kroah.com> wrote:
> > > On Mon, Aug 10, 2009 at 07:18:22PM +0200, Bruno Prémont wrote:
> > > > On Mon, 10 August 2009 Greg KH <greg@kroah.com> wrote:
> > > > > On Mon, Aug 10, 2009 at 11:18:29AM -0400, Alan Stern wrote:
> > > > > > On Sat, 8 Aug 2009, Bruno Prémont wrote:
> > > > > > 
> > > > > > > I tried bisecting this but bisect did end up on a fully
> > > > > > > unrelated commit (which is not even being compiled into my
> > > > > > > kernel). Possibly the failed bisect could be related to
> > > > > > > mis-classified kernel panic/hang while pulling the USB cable
> > > > > > > (there were two such panics for the whole iteration)?
> > > > > > > 
> > > > > > > There are quite a few patches touching tty, ttyUSB and
> > > > > > > friends between rc4 and now so pretty hard to guess on the
> > > > > > > correct one.
> > > > > > > 
> > > > > > > The oops always happens when I disconnect the USB serial
> > > > > > > console (here the one built into Marvell SheevaPlug) while
> > > > > > > having minicom connected to it.
> > > > > > > During the bisection for the last few bad iterations minicom
> > > > > > > got killed (segfault), the bad ones on the iteration left a
> > > > > > > minicom zombie in 'D' state.
> > > 
> > > Ok, a few questions.
> > > 
> > > Are you pulling the device out when you have a console attached to
> > > the device, or just a "normal" minicom connection with it?  I can't
> > > seem to duplicate this here at the moment for some reason :(
> > > 
> > > thanks,
> > > 
> > > greg k-h
> > > 
> > 
> > The SheevaPlug has a getty running on ttyS0 (agetty 115200 ttyS0
> > vt100). That ttyS0 is exported via the FTDI USB to Serial+JTag
> > converter.
> > 
> > On the laptop side (where the kernel Oopses) I see /dev/ttyUSB0 (which
> > I instruct minicom to use via symlink to
> > /dev/serial/by-id/usb-_SheevaPlug_JTAGKey_FT2232D_B_FTS55QK6-if01-port0 -- minicom is
> > limited to 63byte length for path to tty device)
> > 
> > I run minicom with the following parameters:
> >   minicom -o -c on sheeva
> > 
> > /etc/minicom/minirc.sheeva
> >   pu port             /etc/minicom/tty-sheeva
> >   pu rtscts           No
> > 
> > That is 115200 8N1, with neither hardware nor software flow control.
> > 
> > I'm pulling the USB cable while minicom is running and agetty on the
> > other side is waiting for password input. (pulling the USB cable is
> > too easy - and same effect is obtained when pulling power from the
> > SheevaPlug)
> > 
> > 
> > I have not tried yet with a different USB serial device but will do so
> > soon to see if I can reproduce it there. (the other USBSerial device I
> > have is: 067b:2303 Prolific Technology, Inc. PL2303 Serial Port)
> 
> I can't reproduce with this one.
> 
> Similar setup, but instead of SheevaPlug (with it's kernel started with
> console=ttyS0,115200 + agetty run by init) I just started agetty (same
> parameters) on a x86 box with uart serial port (e.g. that x86 kernel
> does not have console kernel parameter) and connected that one via
> NullModem cable to the Prolific USBSerial converter.
> 
> In this case, no matter how often I pull the USBSerial converter there
> is no Oops.
> 
> So it must be something that FTDI driver does differently from e.g.
> Prolific one.

Ok, thanks, I'll dig through my big box of usb devices to try to find a
ftdi device, I was testing this out on a pl2303 device, which is why I
couldn't reproduce it it seems.

greg k-h