Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 15 Sep 2022 18:43:58 +0300
From:   Paul Fertser <fercerpav@gmail.com>
To:     Jiaqing Zhao <jiaqing.zhao@linux.intel.com>
Cc:     Samuel Mendoza-Jonas <sam@mendozajonas.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Paolo Abeni <pabeni@redhat.com>, netdev@vger.kernel.org,
        openbmc@lists.ozlabs.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] net/ncsi: Add Intel OS2BMC OEM command
Message-ID: <YyNIPjNX9MCI3zkK@home.paul.comp>
References: <20220909025716.2610386-1-jiaqing.zhao@linux.intel.com>
 <YxrWPfErV7tKRjyQ@home.paul.comp>
 <8eabb29b-7302-d0a2-5949-d7aa6bc59809@linux.intel.com>
 <Yxrun9LRcFv2QntR@home.paul.comp>
 <36c12486-57d4-c11d-474f-f26a7de8e59a@linux.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <36c12486-57d4-c11d-474f-f26a7de8e59a@linux.intel.com>
Precedence: bulk

Hello,

On Tue, Sep 13, 2022 at 10:12:06AM +0800, Jiaqing Zhao wrote:
> On 2022-09-09 15:43, Paul Fertser wrote:
> > On Fri, Sep 09, 2022 at 03:34:53PM +0800, Jiaqing Zhao wrote:
> >>> Can you please outline some particular use cases for this feature?
> >>>
> >> It enables access between host and BMC when BMC shares the network connection
> >> with host using NCSI, like accessing BMC via HTTP or SSH from host. 
> > 
> > Why having a compile time kernel option here more appropriate than
> > just running something like "/usr/bin/ncsi-netlink --package 0
> > --channel 0 --index 3 --oem-payload 00000157200001" (this example uses
> > another OEM command) on BMC userspace startup?
> > 
> 
> Using ncsi-netlink is one way, but the package and channel id is undetermined
> as it is selected at runtime. Calling the netlink command on a nonexistent
> package/channel may lead to kernel panic.

That sounds like a bug all right. If you can reproduce, it's likely
the fix is reasonably easy, please consider doing it.

> Why I prefer the kernel option is that it applies the config to all ncsi
> devices by default when setting up them. This reduces the effort and keeps
> compatibility. Lots of things in current ncsi kernel driver can be done via
> commands from userspace, but I think it is not a good idea to have a driver
> resides on both kernel and userspace.

How should the developer decide whether to enable this compile-time
option for a platform or not? If it's always nice to have why not
add the code unconditionally? And if not, are you sure kernel compile
time is the right decision point? So far I get an impression a sysfs
runtime knob would be more useful.

-- 
Be free, use free (http://www.gnu.org/philosophy/free-sw.html) software!
mailto:fercerpav@gmail.com