From: "Allen Hubbe" <Allen.Hubbe@dell.com>
To: "'Logan Gunthorpe'" <logang@deltatee.com>,
        <linux-ntb@googlegroups.com>, <linux-pci@vger.kernel.org>,
        <linux-kernel@vger.kernel.org>
Cc: "'Jon Mason'" <jdmason@kudzu.us>,
        "'Dave Jiang'" <dave.jiang@intel.com>,
        "'Bjorn Helgaas'" <bhelgaas@google.com>,
        "'Greg Kroah-Hartman'" <gregkh@linuxfoundation.org>,
        "'Kurt Schwemmer'" <kurt.schwemmer@microsemi.com>,
        "'Stephen Bates'" <sbates@raithlin.com>,
        "'Serge Semin'" <fancer.lancer@gmail.com>
References: <20170629032648.3073-1-logang@deltatee.com> <20170629032648.3073-7-logang@deltatee.com> <000301d2f103$3310ca60$99325f20$@dell.com> <079047c1-6612-2814-e147-844fabd56b75@deltatee.com>
In-Reply-To: <079047c1-6612-2814-e147-844fabd56b75@deltatee.com>
Subject: RE: [PATCH 06/16] ntb: add check and comment for link up to mw_count and mw_get_align
Date: Thu, 29 Jun 2017 16:13:40 -0400
Message-ID: <000401d2f114$3499f2b0$9dcdd810$@dell.com>
MIME-Version: 1.0
Content-Type: text/plain;
        charset="utf-8"
Thread-Index: AQHS8IeVz3JuZyOpkUW5S6r3xXIYA6I78leggACDfwD//8zpcA==
Content-Language: en-us
Sender: linux-kernel-owner@vger.kernel.org
Content-Transfer-Encoding: 8bit
Content-Length: 1491
Lines: 30

From: Logan Gunthorpe
> On 6/29/2017 12:11 PM, Allen Hubbe wrote:
> > Nak.  This breaks a work around for stability issues on some hardware.  I am ok with changing the
> comment to say, this is only supported when called after link up.  I would still like to allow these
> to be called at any time.  Specific hardware drivers like Switchtec may return an error.  Upstream
> drivers, of course, should call these after link up: patch 5/16 part of this set looks good.
> 
> If absolutely necessary I can leave this out. But in terms of interface
> design it's _so_ much better to have it in. This change would bring the
> score from a 3 to a 5 on Rusty Russel's Hard to Misuse ranking[1]. To
> quote Rusty:
> 
> "3. Read the documentation and you'll get it right.
> People only read instructions after they've already tied themselves into
> a knot. Then they skim them for keywords and don't read your warnings. I
> don't give an example of this; if this is the best an interface can get
> do, it's in trouble."
> 
> Can someone not just fix the out-of-tree code? And since when is
> out-of-tree code reasonable justification for what's done in upstream?

Unfortunately, it is to work around hardware errata.  That is not so trivial to fix.

The workaround in software is also not acceptable upstream, for doing things like writing directly to the peer's interrupt controller registers, bypassing the ntb doorbells.

> 
> Logan
> 
> [1]http://ozlabs.org/~rusty/index.cgi/tech/2008-03-30.html