Re: [3.16.1 BISECTED REGRESSION]: Simtec Entropy Key (cdc-acm) broken in 3.16

[Johan Hovold <johan@kernel.org>]

On Thu, Nov 06, 2014 at 01:49:13PM +0000, Nix wrote:
> On 5 Nov 2014, Johan Hovold told this:
> > On Wed, Nov 05, 2014 at 03:14:49PM +0000, Nix wrote:

> > Could you try two more things (too make sure line control is really the
> > culprit):
> >
> > 1. If you clear HUPCL in ekeyd so that the lines are never lowered, does
> > that fix the stability issue?
> 
> Definitely not. I got a hang after the first reboot out of an afflicted
> kernel, when using a HUPCLless ekeyd. Weird. (I guess they're lowered on
> reboot anyway?)

It's actually only the timings related to the control-lines being raised
on open that has changed, so this would seem consistent with that.

> > 2. Can you verify that the patch below works? Did I use the correct
> > VID/PID? Could you provide "lsusb -v" output (for the capability flags)
> > as well?
> 
> The VID/PID are right.
> 
> The patch seems to work. I tested against the usb testing tree directly,
> since that was easier than adjusting it to apply against 3.16. Sixteen
> reboots, no failures, looks to be fixed.

Great, thanks for testing.

> > Note that the patch is against the usb-linus branch of the usb tree,
> 
> OK. (I presume this is gregkh's tree, not yours?)

Yes, Greg's tree, could have been more clear about that.

> > which also contains another fix which could affect this device
> > (set_termios will now be called an extra time on every open). You also
> > need this one, which have not yet been applied:
> >
> > 	http://marc.info/?l=linux-usb&m=141520959813505&w=2
> 
> It had been applied by the time I tested this :)

Greg is fast. :)

Thanks for tracking this down. That bisect cannot have been fun given
the low failure rate (sometimes only one in ten reboots?).

Johan