From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pf0-x233.google.com (mail-pf0-x233.google.com [IPv6:2607:f8b0:400e:c00::233]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 3swPXj4KRTzDt3b for ; Fri, 14 Oct 2016 21:51:17 +1100 (AEDT) Received: by mail-pf0-x233.google.com with SMTP id e6so48366767pfk.3 for ; Fri, 14 Oct 2016 03:51:17 -0700 (PDT) Subject: Re: [PATCH][RFC] powerpc/64s: Use emergency stack for 0x100 system reset interrupt To: Nicholas Piggin , linuxppc-dev@lists.ozlabs.org References: <20161013065106.5389-1-npiggin@gmail.com> Cc: Anton Blanchard From: Balbir Singh Message-ID: <6e4b347e-2c59-ed0a-6012-c8e849f3ff29@gmail.com> Date: Fri, 14 Oct 2016 21:51:16 +1100 MIME-Version: 1.0 In-Reply-To: <20161013065106.5389-1-npiggin@gmail.com> Content-Type: text/plain; charset=windows-1252 List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On 13/10/16 17:51, Nicholas Piggin wrote: > Also use EX_NMI save space. This should give cleaner debug data in > case kernel stacks have become corrupted. > > MSR_RI is kept clear for the 0x100 handler, because we can't mask it, > and can't tolerate a reentrant NMI. Technically the 0x100 is not really > recoverable yet, because it could trample on another emergency stack > user. This will require a bit more work to safely implement, but for > now we don't need to recover after a 0x100. > > A little bit of the emergency stack / MSR_RI handling differences seem > like they can be consolidated with some other handlers. I'll split those > into a couple more patches before submitting. > > I moved the MC save area in the paca out from between the general and > slb save areas, and put it with the nmi area because it shouldn't be used > much. I think we should document the limitations as discussed One of them is that sequential system resets have unpredictable behaviour when done in quick successions. I also don't like some of the naming like NORI, I would rather prefer NOT_RI, but thats just nit picking Balbir Singh.