Re: [REGRESSION] "USB: use kfifo to buffer usb-generic serial writes" causes gobi_loader to hang

[Anssi Hannula <anssi.hannula@iki.fi>]

On lauantai 13 helmikuu 2010 15:35:41 Anssi Hannula wrote:
> On lauantai 13 helmikuu 2010 09:11:36 Oliver Neukum wrote:
> > Am Samstag, 13. Februar 2010 03:50:04 schrieb Alan Stern:
> > > On Sat, 13 Feb 2010, Anssi Hannula wrote:
> > > > On 05.02.2010 23:59, Oliver Neukum wrote:
> > > > > Am Freitag, 5. Februar 2010 20:58:17 schrieb Matthew Garrett:
> > > > >> drivers/usb/serial/usb-serial.c: usb_serial_port_work - port 0
> > > > >> drivers/usb/serial/usb-serial.c: serial_write - port 0, 2048
> > > > >> byte(s) drivers/usb/serial/generic.c: usb_serial_generic_write -
> > > > >> port 0, 2048 bytes drivers/usb/serial/generic.c:
> > > > >> usb_serial_generic_write - put 512 bytes into fifo
> > > > >> drivers/usb/serial/usb-serial.c:
> > > > >> serial_write - port 0, 1536 byte(s) drivers/usb/serial/generic.c:
> > > > >> usb_serial_generic_write - port 0, 1536 bytes
> > > > >> drivers/usb/serial/generic.c: usb_serial_generic_write - put 0
> > > > >> bytes into fifo drivers/usb/serial/generic.c:
> > > > >> usb_serial_generic_write - FIFO is full
> > > > > 
> > > > > OK, could you also get an usbmon trace? This would allow a
> > > > > determination whether the submitted URB doesn't finish for some
> > > > > reason, or whether no URB is submitted, possibly because a wakeup
> > > > > is missed.
> > > > 
> > > > I'm also affected by this regression. Here's an usbmon trace of
> > > > gobi_loader hanging:
> > > > http://stuff.onse.fi/gobi2000/gobi-regression.mon.log
> > > 
> > > That's odd.  The log shows the final bulk-OUT transfer was cancelled
> > > after less than 1 ms.  Is there a timeout value somewhere that is too
> > > small by a factor of 1000?
> > 
> > Neither qcserial nor usb-serial implement timeouts.
> > Is it possible that this is caused by user space closing the handle
> > causing usb-serial::port_release() to call kill_traffic()?
> > 
> > Anssi, going by your log the last write is quite short and begins
> > with 342d3030 39202020 20202020 20202020 801d9b80 02000000 3512dcfe
> > 44313032. Does this match with the last part of the firmware you are
> > transfering?
> 
> Yes. (there are three files, this is the last part of the third file)
> 
> > Could you put an "mdelay(500);" at the beginning of
> > usb-serial::port_release() and retest?
> 
> It seems I can't get that far anymore, it hangs much more early now:
> http://stuff.onse.fi/gobi2000/gobi-regression2.mon.log
> After interrupting gobi_loader I get a WARNING:
> http://stuff.onse.fi/gobi2000/gobi-regression2.warning.log
> Trying gobi_loader again immediately, the log is almost empty:
> http://stuff.onse.fi/gobi2000/gobi-regression3.mon.log
> After reboot I get a similar log as regression2:
> http://stuff.onse.fi/gobi2000/gobi-regression4.mon.log
> 
> Actually, just now I noticed that the first time I had some extra sleep
> calls before starting the firmware upload in gobi_loader, which caused it
> to successfully finish (i.e. not hang; I guess somehow I didn't notice it
> when I logged the first usbmon.log), but the firmware wasn't still
> uploaded properly (that'd be because of the cancelled transfer I guess).
> 
> I'll get back to this later today, including testing with mdelay in
> port_release with the extra sleep calls in gobi_loader.

port_release is only called when the serial device is removed, so this has no 
effect here.

-- 
Anssi Hannula