From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from caramon.arm.linux.org.uk ([217.147.92.249]:49356 "EHLO caramon.arm.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757827AbXGFVzg (ORCPT ); Fri, 6 Jul 2007 17:55:36 -0400 Date: Fri, 6 Jul 2007 22:54:44 +0100 From: Russell King Subject: Re: + print-utsname-on-oops-on-all-architectures.patch added to -mm tree Message-ID: <20070706215444.GA1991@flint.arm.linux.org.uk> References: <200707061754.l66HsXfb010994@imap1.linux-foundation.org> <20070706182612.GA26179@flint.arm.linux.org.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Sender: linux-arch-owner@vger.kernel.org To: Joshua Wise Cc: akpm@linux-foundation.org, mm-commits@vger.kernel.org, ak@suse.de, benh@kernel.crashing.org, ink@jurassic.park.msu.ru, linux-arch@vger.kernel.org, paulus@samba.org, rth@twiddle.net List-ID: On Fri, Jul 06, 2007 at 02:03:45PM -0700, Joshua Wise wrote: > On Fri, 6 Jul 2007, Russell King wrote: > > On Fri, Jul 06, 2007 at 10:54:33AM -0700, akpm@linux-foundation.org wrote: > > > Potential impact: > > > This adds another line to the Oops output, causing the first few lines to > > > potentially scroll off the screen. This also adds a few more pointer > > > dereferences in the Oops path, because it adds to the die_chain notifier > > > chain, reducing the likelihood that the Oops will be printed if there is > > > very bad memory corruption. > > > > Plus we don't get the utsname information on oops dumps during the kernel > > initialisation. Not good - I'd rather keep things as is rather than loose > > that facility which we've just gained on ARM. > > Ok -- would it be more agreeable to make that a core_initcall? That just makes the reporting slightly earlier and doesn't solve the real problem. > > Instead, can we have this as a call-able function which returns the > > pointer, or just make the pointer global. That way we can also eliminate > > the need for another additional line in the oops output. > > Hm. I don't really like the precedent being set for sucking arbitrary > globals into the Oops message. It's much safer than walking notifier lists and such like. > In particular, I am also considering a patch to print machine check > statistics (if we took any machine check exceptions) at Oops-time, and I > would implement that in a similar manner to this. I'm also going to raise an additional objection - whoever introduced notify_die() hasn't bothered fixing up the ARM implementation, so your patch actually _removes_ functionality recently introduced. In addition, I notice that the introduction of notify_die() didn't take account of what other architectures do (eg passing siginfo structures), so to convert ARM to using it would either require us to drop some functionality or re-engineer notify_die(). Therefore, I'm going to ask you to remove the changes from your patch which touch the ARM architecture since the overall effect is a net removal of functionality I've only recently merged. -- Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: