svn commit: r195960 - in head/sys/dev/usb: . controller input

[Hans Petter Selasky]

On Friday 07 August 2009 08:54:00 M. Warner Losh wrote:
> In message: <200908070830.47894.hselasky at c2i.net>
>
>             Hans Petter Selasky <hselasky at c2i.net> writes:
> : On Thursday 06 August 2009 21:47:16 Navdeep Parhar wrote:
> : > >> See attached patch. Please test and report back.
> : > >
> : > > This patch fixes my problem.  The machine is remote and I'm unable
> : > > to test whether the USB keyboard and keystroke repetition works, but
> : > > core dumps to a SATA disk are now as fast as they were before
> : > > r195960.  Thanks.
> : >
> : > I finally got a chance to try a USB keyboard with ddb, and things did
> : > not go too well overall.  While inside ddb, keystrokes were recognized
> : > properly and repetition worked too.  But after exiting ddb, the
> : > keyboard wouldn't work - there wasn't any visible response to
> : > keystrokes.  Also, I kept seeing the login prompt continually scroll
> : > up, as if someone was pressing <return> repeatedly.  It certainly
> : > wasn't me :-)
> : >
> : > Are you assuming that a user will not resume normal operation after
> : > entering the debugger?  A panic/reboot isn't the only exit route from
> : > ddb.....
> : >
> : > Simple sequence of steps to reproduce problem:
> : > ctrl-alt-esc on the USB keyboard
> : > db> c<return>
> :
> : This is like expected.
> :
> : Once paniced, USB operation is blocked on the USB controller which the
> : keyboard belongs to, because USB does not receive any polling-complete
> : call, so that it can clean up the state in the USB controller! This
> : mainly has to do with avoid calling wakeup() during polling.
> :
> : To avoid wakeup() calls, USB sets some bits, which must be cleared when
> : polling is complete, which is currently not done, because USB doesn't
> : know when polling is complete ...
>
> Polling isn't supposed to work like this...  The rest of the system
> effects a poll without these side effects.

Please try the following patch and report back. There is some Giant fuzz which 
you can ignore. I will see if I can followup a patch against -current later 
today.

http://perforce.freebsd.org/chv.cgi?CH=167084

--HPS