[trsqr]

We will find the culprit tonight, tsqr has started bisecting and 4 steps are left. Stay tuned.

Ok, I've compiled 7 kernels now with some serious git+kernel compilation guidance from fritsch. As a result it seems that this commit is the one causing the issue. It seems that this same commit breaks multiple PCTV 290e tuners in one NUC as well.

Code:

47f467ac740ebf0475a5176ddb1741acba6aad4e is the first bad commit
commit 47f467ac740ebf0475a5176ddb1741acba6aad4e
Author: Sarah Sharp <[email protected]>
Date:   Fri Jan 31 11:45:02 2014 -0800

    Revert "xhci: Set scatter-gather limit to avoid failed block writes."

    commit 1386ff75797a187df324062fb4e929152392da88 upstream.

    This reverts commit f2d9b991c549f159dc9ae81f77d8206c790cbfee.

    We are ripping out commit 35773dac5f862cb1c82ea151eba3e2f6de51ec3e "usb:
    xhci: Link TRB must not occur within a USB payload burst" because it's a
    hack that caused regressions in the usb-storage and userspace USB
    drivers that use usbfs and libusb.  This commit attempted to fix the
    issues with that patch.

    Signed-off-by: Sarah Sharp <[email protected]>
    Signed-off-by: Greg Kroah-Hartman <[email protected]>

:040000 040000 632634f31130502e87cd3e5fd0e300d5a8249ffd 8f1de710a99ac9b721bb77dc0dd2ab9de9597029 M       drivers

The good ones were very stable. I recorded 2 HD channels, watched 1 other HD channel in XBMC with EPG overlaid on it and used another tuner to record SD channel while compiling the next kernel, and not a single continuity error was seen during the compilation.

This was an eye-opening experience. Nice that we masterminds like fritsch who are able to come up with such a test setup, that a lesser Linux mortal like myself could actually make the tests and contribute in some way. Now I would need suggestions on how to proceed with the bug reporting.

Edit: The way I tested was: started with a single tuner connected. Started to record a HDTV channel and watch another one in XBMC. Then tried to watch the EPG and move around in the menus. In some of the versions this caused continuity counter errors in the logs. Then I enabled the second tuner. For the softwares that I saw continuity counter errors, there was also an XHCI crash when the second tuner was used. So it seems that the root cause for both problems is the same.