From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Message-Id: From: Kumar Gala To: Paul Gortmaker In-Reply-To: <20080108052743.GA18083@windriver.com> Content-Type: text/plain; charset=US-ASCII; format=flowed; delsp=yes Mime-Version: 1.0 (Apple Message framework v915) Subject: Re: [PATCH/RFC] mpc83xx/85xx SysRQ/brk over 8250 console Date: Tue, 8 Jan 2008 00:11:19 -0600 References: <20080108052743.GA18083@windriver.com> Cc: Linuxppc-dev@ozlabs.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Jan 7, 2008, at 11:27 PM, Paul Gortmaker wrote: > It seems that if a break is rec'd on the serial console (as per SysRQ > or similar) on some 83xx and 85xx processors, then a pulse of IRQs is > generated which makes the use of SysRQ itself about 99% impossible. I > experimented with trying several ACK strategies, but in the end the > thing which worked best was just to wait for the surge of events to > pass > (a fraction of a second). The number of events can be on the order of > 1000 (as reported by /proc/interrupts) > > I really dislike seeing board specific ifdefs within what should be > board independent code, so I'm hoping that once the problem is known, > that a more elegant solution will pop into someone's head. Or at > least > this will hopefully save someone the grief of re-investigating the > source > and start a discussion. > > I've seen this on two different 834x boards and also an 8548 based > board. At this point, it is not clear to me if the problem extends > beyond these CPU/soc to all with UARTs at 4500/4600. > > While less than ideal, the work-around below which simply ignores the > events does allow a person to use SysRQ on such platforms. > > The definition of the RFE bit (Rx FIFO error) can be found in pretty > much any of the MPC CPU PDFs (for those with 4500/4600 16550s). > > Paul. This should really get posted to linux-kernel@vger.kernel.org and probably linux-serial@vger.kernel.org lists. - k