Re: [PATCH net-next 0/6] mlxsw: Add extended ACK for EMADs

From: Jakub Kicinski <jakub.kicinski@netronome.com>
To: Jiri Pirko <jiri@resnulli.us>
Cc: Ido Schimmel <idosch@idosch.org>,
	netdev@vger.kernel.org, davem@davemloft.net, jiri@mellanox.com,
	shalomt@mellanox.com, mlxsw@mellanox.com,
	Ido Schimmel <idosch@mellanox.com>
Subject: Re: [PATCH net-next 0/6] mlxsw: Add extended ACK for EMADs
Date: Wed, 6 Nov 2019 09:26:47 -0800	[thread overview]
Message-ID: <20191106092647.4de42312@cakuba.netronome.com> (raw)
In-Reply-To: <20191106082039.GB2112@nanopsycho>

On Wed, 6 Nov 2019 09:20:39 +0100, Jiri Pirko wrote:
> Tue, Nov 05, 2019 at 10:48:58PM CET, jakub.kicinski@netronome.com wrote:
> >On Tue, 5 Nov 2019 22:48:26 +0200, Ido Schimmel wrote:  
> >> On Tue, Nov 05, 2019 at 09:54:48AM -0800, Jakub Kicinski wrote:  
> >> > Hm, the firmware has no log that it keeps? Surely FW runs a lot of
> >> > periodic jobs etc which may encounter some error conditions, how do 
> >> > you deal with those?    
> >> 
> >> There are intrusive out-of-tree modules that can get this information.
> >> It's currently not possible to retrieve this information from the
> >> driver. We try to move away from such methods, but it can't happen
> >> overnight. This set and the work done in the firmware team to add this
> >> new TLV is one step towards that goal.
> >>   
> >> > Bottom line is I don't like when data from FW is just blindly passed
> >> > to user space.    
> >> 
> >> The same information will be passed to user space regardless if you use
> >> ethtool / devlink / printk.  
> >
> >Sure, but the additional hoop to jump through makes it clear that this
> >is discouraged and it keeps clear separation between the Linux
> >interfaces and proprietary custom FW.. "stuff".  
> 
> Hmm, let me try to understand. So you basically have problem with
> passing random FW generated data and putting it to user (via dmesg,
> extack). However, ethtool dump is fine. Devlink health reporter is also
> fine.

Yup.

> That is completely sufficient for async events/errors.
> However in this case, we have MSG sent to fw which generates an ERROR
> and this error is sent from FW back, as a reaction to this particular
> message.

Well, outputting to dmesg is not more synchronous than putting it in
some other device specific facility.

> What do you suggest we should use in order to maintain the MSG-ERROR
> pairing? Perhaps a separate devlink health reporter just for this?

Again, to be clear - that's future work, right? Kernel logs as
implemented here do not maintain MSG-ERROR pairing.

> What do you think?

In all honesty it's hard to tell for sure, because we don't see the FW
and what it needs. That's kind of the point, it's a black box.

I prefer the driver to mediate all the information in a meaningful way.
If you want to report an error regarding a parameter the FW could
communicate some identification of the field and the reason and the
driver can control the output, for example format a string to print to
logs?