Re: [PATCH 3/5] nmi: add MCE class for implementing machine check injection commands

From: David Gibson <david@gibson.dropbear.id.au>
To: Nicholas Piggin <npiggin@gmail.com>
Cc: "Alexey Kardashevskiy" <aik@ozlabs.ru>,
	"Mahesh Salgaonkar" <mahesh@linux.vnet.ibm.com>,
	"Greg Kurz" <groug@kaod.org>,
	qemu-devel@nongnu.org, "Cédric Le Goater" <clg@fr.ibm.com>,
	"Ganesh Goudar" <ganeshgr@linux.ibm.com>,
	qemu-ppc@nongnu.org
Subject: Re: [PATCH 3/5] nmi: add MCE class for implementing machine check injection commands
Date: Mon, 6 Apr 2020 16:45:59 +1000	[thread overview]
Message-ID: <20200406064559.GB2945@umbus.fritz.box> (raw)
In-Reply-To: <1585900632.k0ft0cc80l.astroid@bobo.none>

[-- Attachment #1: Type: text/plain, Size: 2227 bytes --]

On Fri, Apr 03, 2020 at 06:04:47PM +1000, Nicholas Piggin wrote:
> David Gibson's on March 31, 2020 10:22 am:
> > On Thu, Mar 26, 2020 at 12:41:45AM +1000, Nicholas Piggin wrote:
> >> Like commit 9cb805fd26 ("cpus: Define callback for QEMU "nmi" command")
> >> this implements a machine check injection command framework and defines
> >> a monitor command for ppc.
> >> 
> >> Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> > 
> > So, AFAICT, both x86 and ppc have something called an MCE, and while
> > I'm guessing they're somewhat related, they don't work quite the same
> > way - different information provided to the handler and so forth.
> > 
> > I think it's reasonable to overload the "mce" HMP command based on
> > target for the different types.  However, I think calling the
> > interface classes which are specific to the ppc type just "mce" could
> > be pretty confusing.
> 
> Okay. So, convert i386 first?

Uh.. not necessarily.  But call your version PpcMce or something
instead.

> > In addition, I think this is adding an HMP command to inject the event
> > without any corresponding way of injecting via QMP.  I believe that's
> > frowned upon.
> 
> I attempted that but didn't get too far. I guess it's more of a
> special test than a management function (nmi has valid uses in 
> administering a machine), so maybe we can get an exemption. One issue
> is different QMP command for powerpc vs x86.
> 
> I think error injection as a general concept might be valid there, but
> the better interface for that level would be higher up, e.g, not
> specifying register settings but rather "simulate uncorrected memory
> error on this byte".
> 
> Do you think that is reasonable reason to avoid adding QMP for this
> nasty low level thing?

Hrm, I doubt it will convince the qemu community at large.  I think
the view is that QMP is the "real" management interface for qemu - HMP
is somewhere in between a shim for convenience and something needed
for backwards compat.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]