Re: [PATCH net-next v16 3/3] net: ax88796c: ASIX AX88796C SPI Ethernet Adapter Driver

From: Lukasz Stelmach <l.stelmach@samsung.com>
To: Jakub Kicinski <kuba@kernel.org>
Cc: "David S. Miller" <davem@davemloft.net>,
	"Andrew Lunn" <andrew@lunn.ch>,
	jim.cromie@gmail.com, "Heiner Kallweit" <hkallweit1@gmail.com>,
	"Rob Herring" <robh+dt@kernel.org>,
	"Kukjin Kim" <kgene@kernel.org>,
	"Krzysztof Kozlowski" <krzk@kernel.org>,
	"Russell King" <linux@armlinux.org.uk>,
	netdev@vger.kernel.org, devicetree@vger.kernel.org,
	linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org,
	linux-samsung-soc@vger.kernel.org,
	"Bartłomiej Żolnierkiewicz" <b.zolnierkie@samsung.com>,
	"Marek Szyprowski" <m.szyprowski@samsung.com>
Subject: Re: [PATCH net-next v16 3/3] net: ax88796c: ASIX AX88796C SPI Ethernet Adapter Driver
Date: Wed, 13 Oct 2021 23:14:20 +0200	[thread overview]
Message-ID: <dleftj1r4osk43.fsf%l.stelmach@samsung.com> (raw)
In-Reply-To: <20211001153940.28c5d60d@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com> (Jakub Kicinski's message of "Fri, 1 Oct 2021 15:39:40 -0700")

[-- Attachment #1: Type: text/plain, Size: 2482 bytes --]

It was <2021-10-01 pią 15:39>, when Jakub Kicinski wrote:
> On Wed, 29 Sep 2021 16:08:54 +0200 Łukasz Stelmach wrote:
>> +static char *no_regs_list = "80018001,e1918001,8001a001,fc0d0000";
>
> static const char ...
>
>> +static int
>> +ax88796c_close(struct net_device *ndev)
>> +{
>> +	struct ax88796c_device *ax_local = to_ax88796c_device(ndev);
>> +
>> +	netif_stop_queue(ndev);
>
> This can run concurrently with the work which restarts the queue.
> You should take the mutex and purge the queue here, so that there 
> is no chance queue will get restarted by the work right after.
>> +	phy_stop(ndev->phydev);
>> +
>> +	mutex_lock(&ax_local->spi_lock);

This was a bit tricky. The function now looks like this (details in the
comments). In theory there is a chance of dropping a packet if the
ax88796c_work() runs between phy_stop() and mutex_lock(), but I'd say
the overall risk is negligible.

--8<---------------cut here---------------start------------->8---
static int
ax88796c_close(struct net_device *ndev)
{
        struct ax88796c_device *ax_local = to_ax88796c_device(ndev);

        phy_stop(ndev->phydev);

        /* We lock the mutex early not only to protect the device
         * against concurrent access, but also avoid waking up the
         * queue in ax88796c_work(). phy_stop() needs to be called
         * before because it locks the mutex to access SPI.
         */
        mutex_lock(&ax_local->spi_lock);

        netif_stop_queue(ndev);

        /* No more work can be scheduled now. Make any pending work,
         * including one already waiting for the mutex to be unlocked,
         * NOP.
         */
        clear_bit(EVENT_SET_MULTI, &ax_local->flags);
        clear_bit(EVENT_INTR, &ax_local->flags);
        clear_bit(EVENT_TX, &ax_local->flags);

        /* Disable MAC interrupts */
        AX_WRITE(&ax_local->ax_spi, IMR_MASKALL, P0_IMR);
        __skb_queue_purge(&ax_local->tx_wait_q);
        ax88796c_soft_reset(ax_local);

        mutex_unlock(&ax_local->spi_lock);

        cancel_work_sync(&ax_local->ax_work);

        free_irq(ndev->irq, ndev);

        return 0;
}
--8<---------------cut here---------------end--------------->8---

>> +MODULE_AUTHOR("ASIX");
>
> You can drop this, author should be human.

I looked at diff --stat and decided it is OK to put my name there (-;
-- 
Łukasz Stelmach
Samsung R&D Institute Poland
Samsung Electronics

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]