From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from gate.crashing.org (gate.crashing.org [63.228.1.57]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client did not present a certificate) by ozlabs.org (Postfix) with ESMTP id 2668F679A6 for ; Tue, 18 Apr 2006 16:00:44 +1000 (EST) Subject: Re: 7447A strange problem with MSR:POW (WAS: can't boot 2.6.17-rc1) From: Benjamin Herrenschmidt To: Paul Mackerras In-Reply-To: <17476.31998.168122.413195@cargo.ozlabs.ibm.com> References: <1144408805.30891.42.camel@localhost.localdomain> <17463.9759.442768.685153@cargo.ozlabs.ibm.com> <1144923633.4935.11.camel@localhost.localdomain> <21F7D7D8-B9BC-44EB-B07B-F888D89DCF25@freescale.com> <17471.62187.774127.783000@cargo.ozlabs.ibm.com> <1145048461.4223.38.camel@localhost.localdomain> <17476.31998.168122.413195@cargo.ozlabs.ibm.com> Content-Type: text/plain Date: Tue, 18 Apr 2006 16:00:34 +1000 Message-Id: <1145340034.4705.55.camel@localhost.localdomain> Mime-Version: 1.0 Cc: Becky Bruce , Michael Schmitz , debian-powerpc@lists.debian.org, linuxppc-dev list List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Tue, 2006-04-18 at 15:45 +1000, Paul Mackerras wrote: > Benjamin Herrenschmidt writes: > > > Looks good to me except that we need the same for ppc64 since the 970 > > theorically has the same problem... > > OK, does this look OK to everyone, before I send it off to Linus? I > now use a bit in the thread_info rather than using the HID0 bits > themselves to indicate that we're napping, since the m[ft]spr might be > slow. I added a `local_flags' field to the thread_info struct for > things that are only changed by the task itself and therefore don't > need to be accessed atomically. > > This version does the same sort of change for the 970 as for 6xx. Hrm... The 970 version bloats the exception prolog significantly... I understand now why you were talking about putting the code in the exit path on irc ... I don't like it that way.... Also, if you want to keep it, maybe use a separate CONFIG_PPC_970STYLE_NAP or something that gets selected by platforms that can do it ? I suppose a PACA field would be less inefficient but still sucks... the exception return to userland code path already accesses thread_info and definitely looks like a better place to put it... as long as we never have to add dodgy workarounds when getting out of NAP like we do on 6xx. Ben.