Re: [PATCH net] net: dsa: call teardown method on probe failure

From: Vladimir Oltean <vladimir.oltean@nxp.com>
To: Andrew Lunn <andrew@lunn.ch>
Cc: "David S . Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, netdev <netdev@vger.kernel.org>,
	Florian Fainelli <f.fainelli@gmail.com>,
	Vivien Didelot <vivien.didelot@gmail.com>
Subject: Re: [PATCH net] net: dsa: call teardown method on probe failure
Date: Thu, 4 Feb 2021 17:06:15 +0000	[thread overview]
Message-ID: <20210204170614.zutxxuufsx53lcgg@skbuf> (raw)
In-Reply-To: <YBwoKiRlOmi3my5G@lunn.ch>

On Thu, Feb 04, 2021 at 06:00:26PM +0100, Andrew Lunn wrote:
> On Thu, Feb 04, 2021 at 06:33:51PM +0200, Vladimir Oltean wrote:
> > Since teardown is supposed to undo the effects of the setup method, it
> > should be called in the error path for dsa_switch_setup, not just in
> > dsa_switch_teardown.
> 
> I disagree with this. If setup failed, it should of cleaned itself up.
> That is the generally accepted way of doing things. If a function is
> going to exit with an error, it should first undo whatever it did
> before exiting.
> 
> You are adding extra semantics to the teardown op. It can no longer
> assume setup was successful. So it needs to be very careful about what
> it tears down, it cannot assume everything has been setup. I doubt the
> existing implementations actually do that.

I'm sorry, I don't understand.
I write a driver, I implement .setup(). I allocate some memory, I expect
that I can deallocate it in .teardown().
Now dsa_switch_setup comes, calls my .setup() which succedes. But then
mdiobus_register(ds->slave_mii_bus) which comes right after .setup()
fails. Are you saying we shouldn't call the driver's .teardown()?
Why not?