linux-mips
[Top] [All Lists]

Re: [PATCH 8/12] cerr-printk-not-prom-printf

To: Andrew Isaacson <adi@broadcom.com>
Subject: Re: [PATCH 8/12] cerr-printk-not-prom-printf
From: Ralf Baechle <ralf@linux-mips.org>
Date: Thu, 20 Oct 2005 11:49:02 +0100
Cc: linux-mips@linux-mips.org
In-reply-to: <20051020065757.GH23899@broadcom.com>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <20051020065320.GA23857@broadcom.com> <20051020065757.GH23899@broadcom.com>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mutt/1.4.2.1i
On Wed, Oct 19, 2005 at 11:57:57PM -0700, Andrew Isaacson wrote:

> Use printk rather than prom_printf in cache error handlers.
> prom_printf does not handle SMP locking, and shows up only
> on console (not in dmesg(1) or syslog).

This patch essentially reverses

  
http://www.linux-mips.org/git?p=linux.git;a=blobdiff;h=b2d41a2b0c406db6be87d02f51600e7e3013a748;hp=051290842231d9734d0bb2dc4c56b8590c4f5337;hb=760190d3f99ebabd965bd35324d5d084ff266711;f=arch/mips/mm/cerr-sb1.c

(Yuck, gitk URLs are bloody long ...)

The reason for this old commit was that this code is running uncached, so
the operation of ll/sc in the spinlocks is undefined according to the
MIPS64 spec, also there is a potencial deadlock since printk's
logbuf_lock might be held.

The old pass 1 part did truely exercise the cache error handler ;-)

  Ralf

<Prev in Thread] Current Thread [Next in Thread>