Re: [PATCH v3 0/4] Regression fixes for rfcomm/tty.c

Am 28.01.2014 09:31, schrieb Gianluca Anzolin:

Unfortunately it seems I overlooked the fact that rfcomm_dev_activate() is
called with the port->mutex held. So patches 2/3/4 cause a regression I missed
because I didn't turn on the appropriate debug options (circular locking
dependency, a bug report already appeared on this list).

I'm afraid this all stems from my partial knowledge of the tty_port code and
unfortunately I don't know how to solve the problem right now.

Understandable, it's like a mine field. ;) Maybe it might make sense to add Alan Cox to Cc, I think he's active again and knows a lot about tty_port.

I think it's better to revert those patches for the moment.

I prefer to still use those patches because without them, I have a more serious problem (at least for my use cases, which happily haven't run into that deadlock).

But thanks for notifying me/us about the possibility of a deadlock when using your patches.

Regards,

Alexander Holler