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:   Wed, 4 Jan 2023 19:32:53 +0200
From:   Vladimir Oltean <olteanv@gmail.com>
To:     "Hans J. Schultz" <netdev@kapio-technology.com>
Cc:     davem@davemloft.net, kuba@kernel.org, netdev@vger.kernel.org,
        Andrew Lunn <andrew@lunn.ch>,
        Florian Fainelli <f.fainelli@gmail.com>,
        Eric Dumazet <edumazet@google.com>,
        Paolo Abeni <pabeni@redhat.com>,
        open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v2 net-next 2/3] net: dsa: mv88e6xxx: disable hold of
 chip lock for handling
Message-ID: <20230104173253.77wa6kmi4fzglc6v@skbuf>
References: <20230104130603.1624945-1-netdev@kapio-technology.com>
 <20230104130603.1624945-1-netdev@kapio-technology.com>
 <20230104130603.1624945-3-netdev@kapio-technology.com>
 <20230104130603.1624945-3-netdev@kapio-technology.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230104130603.1624945-3-netdev@kapio-technology.com>
 <20230104130603.1624945-3-netdev@kapio-technology.com>
Precedence: bulk

On Wed, Jan 04, 2023 at 02:06:02PM +0100, Hans J. Schultz wrote:
> As functions called under the interrupt handler will need to take the
> netlink lock, we need to release the chip lock before calling those
> functions as otherwise double lock deadlocks will occur as userspace
> calls towards the driver often take the netlink lock and then the
> chip lock.
> 
> The deadlock would look like:
> 
> Interrupt handler: chip lock taken, but cannot take netlink lock as
>                    userspace config call has netlink lock.
> Userspace config: netlink lock taken, but cannot take chip lock as
>                    the interrupt handler has the chip lock.

Ultimately, none of this explanation is really relevant, and it requires
too much prior reviewer knowledge. I would phrase the commit title as
"shorten the locked section in mv88e6xxx_g1_atu_prob_irq_thread_fn()"
and say, as an explanation, that only the hardware access functions (up
until the call to mv88e6xxx_g1_atu_mac_read()) require the register lock.
The follow-up code, which processes the ATU violation data, can run
unlocked, and in a future patch will even run from a context which is
incompatible with the register lock being held. If you wish, you can
mention here as a small note that the incompatible context comes from an
AB/BA ordering inversion with rtnl_lock().

> 
> Signed-off-by: Hans J. Schultz <netdev@kapio-technology.com>
> ---
>  drivers/net/dsa/mv88e6xxx/global1_atu.c | 14 ++++++++------
>  1 file changed, 8 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/dsa/mv88e6xxx/global1_atu.c b/drivers/net/dsa/mv88e6xxx/global1_atu.c
> index 61ae2d61e25c..34203e112eef 100644
> --- a/drivers/net/dsa/mv88e6xxx/global1_atu.c
> +++ b/drivers/net/dsa/mv88e6xxx/global1_atu.c
> @@ -409,11 +409,11 @@ static irqreturn_t mv88e6xxx_g1_atu_prob_irq_thread_fn(int irq, void *dev_id)
>  
>  	err = mv88e6xxx_g1_read_atu_violation(chip);
>  	if (err)
> -		goto out;
> +		goto out_unlock;
>  
>  	err = mv88e6xxx_g1_read(chip, MV88E6XXX_G1_ATU_OP, &val);
>  	if (err)
> -		goto out;
> +		goto out_unlock;
>  
>  	err = mv88e6xxx_g1_atu_fid_read(chip, &fid);

If mv88e6xxx_g1_atu_fid_read() fails, it will goto out, which will exit
the IRQ handler with the mv88e6xxx_reg_lock() still held.

Probably not a good idea, since the driver will access the registers
again in the future (errors in IRQ handlers aren't propagated anywhere),
and the user might need a computer which is not deadlocked.

>  	if (err)
> @@ -421,11 +421,13 @@ static irqreturn_t mv88e6xxx_g1_atu_prob_irq_thread_fn(int irq, void *dev_id)
>  
>  	err = mv88e6xxx_g1_atu_data_read(chip, &entry);
>  	if (err)
> -		goto out;
> +		goto out_unlock;
>  
>  	err = mv88e6xxx_g1_atu_mac_read(chip, &entry);
>  	if (err)
> -		goto out;
> +		goto out_unlock;
> +
> +	mv88e6xxx_reg_unlock(chip);
>  
>  	spid = entry.state;
>  
> @@ -449,13 +451,13 @@ static irqreturn_t mv88e6xxx_g1_atu_prob_irq_thread_fn(int irq, void *dev_id)
>  						   fid);
>  		chip->ports[spid].atu_full_violation++;
>  	}
> -	mv88e6xxx_reg_unlock(chip);
>  
>  	return IRQ_HANDLED;
>  
> -out:
> +out_unlock:
>  	mv88e6xxx_reg_unlock(chip);
>  
> +out:
>  	dev_err(chip->dev, "ATU problem: error %d while handling interrupt\n",
>  		err);
>  	return IRQ_HANDLED;
> -- 
> 2.34.1
>