From mboxrd@z Thu Jan 1 00:00:00 1970 From: Peter Hurley Subject: Re: 3.8-rc regression with pps-ldisc due to 70ece7a731 Date: Wed, 06 Feb 2013 10:53:40 -0500 Message-ID: <1360166020.3298.42.camel@thor.lan> References: <20130204010303.24971.qmail@science.horizon.com> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from mailout39.mail01.mtsvc.net ([216.70.64.83]:34006 "EHLO n12.mail01.mtsvc.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1756394Ab3BFPxv (ORCPT ); Wed, 6 Feb 2013 10:53:51 -0500 In-Reply-To: <20130204010303.24971.qmail@science.horizon.com> Sender: linux-serial-owner@vger.kernel.org List-Id: linux-serial@vger.kernel.org To: George Spelvin Cc: jslaby@suse.cz, linux-serial@vger.kernel.org, linux-kernel@vger.kernel.org Hi George, On Sun, 2013-02-03 at 20:03 -0500, George Spelvin wrote: > "TTY: n_tty, add ldisc data to n_tty" > > The PPS line discipline has incestuous relations with the n_tty line > discipline, using some hooks to call internal routines. > > However, I started noticing violent kernel explosions when testing 3.8-rc, > and after a bit of digging, I think it's due to the fact that the PPS code > assumes that the ->ldisc pointer is available to hold a pointer to a "pps" > structure, but this commit started using it in the core n_tty discipline. > > If you look at pps_tty_open in drivers/pps/clients/pps-ldisc.c, you can > see it does: > > pps = pps_register_source(&info, PPS_CAPTUREBOTH | \ > PPS_OFFSETASSERT | PPS_OFFSETCLEAR); > if (pps == NULL) { > pr_err("cannot register PPS source \"%s\"\n", info.path); > return -ENOMEM; > } > tty->disc_data = pps; > > /* Should open N_TTY ldisc too */ > ret = alias_n_tty_open(tty); > > Where "alias_n_tty_open" is filled in by n_tty_inherit_ops() to be > n_tty_open(). However, in this commit, n_tty_open() now allocates > its own structure and overwrites the disc_data pointer, leading to an > earth-shattering kaboom as NULL pointers are dereferenced in interrupt > handlers. Yuck. Little wonder it broke. > My first thought is to reserve a pointer in n_tty_data for the > pps structure and update the pps code to chase pointers one more > level. But I wanted to solicit opinions. Tight coupling is what caused this to break in the first place -- I don't think tighter coupling is the right answer. Alternate coming... Regards, Peter Hurley