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, 16 Nov 2022 11:27:58 +0100
From:   Francois Romieu <romieu@fr.zoreil.com>
To:     Zhang Changzhong <zhangchangzhong@huawei.com>
Cc:     davem@davemloft.net, edumazet@google.com, kuba@kernel.org,
        pabeni@redhat.com, mdf@kernel.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH net v2 3/3] net: nixge: fix tx queue handling
Message-ID: <Y3S7LpbVaTRdKmLl@electric-eye.fr.zoreil.com>
References: <1668525024-38409-1-git-send-email-zhangchangzhong@huawei.com>
 <1668525024-38409-4-git-send-email-zhangchangzhong@huawei.com>
 <Y3Qa/fjjMhctsE5w@electric-eye.fr.zoreil.com>
 <c476086a-14ce-6e47-8183-def31d569ec6@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <c476086a-14ce-6e47-8183-def31d569ec6@huawei.com>
Precedence: bulk

Zhang Changzhong <zhangchangzhong@huawei.com> :
> On 2022/11/16 7:04, Francois Romieu wrote:
> > Zhang Changzhong <zhangchangzhong@huawei.com> :
[...]
> >> diff --git a/drivers/net/ethernet/ni/nixge.c b/drivers/net/ethernet/ni/nixge.c
> >> index 91b7ebc..3776a03 100644
> >> --- a/drivers/net/ethernet/ni/nixge.c
> >> +++ b/drivers/net/ethernet/ni/nixge.c
> > [...]
> >>  static netdev_tx_t nixge_start_xmit(struct sk_buff *skb,
> >> @@ -518,10 +523,15 @@ static netdev_tx_t nixge_start_xmit(struct sk_buff *skb,
> >>  	cur_p = &priv->tx_bd_v[priv->tx_bd_tail];
> >>  	tx_skb = &priv->tx_skb[priv->tx_bd_tail];
> >>  
> >> -	if (nixge_check_tx_bd_space(priv, num_frag + 1)) {
> >> -		if (!netif_queue_stopped(ndev))
> >> -			netif_stop_queue(ndev);
> >> -		return NETDEV_TX_OK;
> >> +	if (unlikely(nixge_check_tx_bd_space(priv, num_frag + 1))) {
> >> +		/* Should not happen as last start_xmit call should have
> >> +		 * checked for sufficient space and queue should only be
> >> +		 * woken when sufficient space is available.
> >> +		 */
> > 
> > Almost. IRQ triggering after nixge_start_xmit::netif_stop_queue and
> > before nixge_start_xmit::smp_mb may wrongly wake queue.
> > 
> 
> I don't know what you mean by "wronly wake queue". The queue is woken
> only when there is sufficient for next packet.

Between nixge_start_xmit::netif_stop_queue and nixge_start_xmit::smp_mb,
"next" packet is current packet in hard_start_xmit. However said current
packet may not be accounted for in the IRQ context transmit completion
handler.

[nixge_start_xmit]

        ++priv->tx_bd_tail;
        priv->tx_bd_tail %= TX_BD_NUM;

	/* Stop queue if next transmit may not have space */
	if (nixge_check_tx_bd_space(priv, MAX_SKB_FRAGS + 1)) {
                netif_stop_queue(ndev);

Which value does [nixge_start_xmit_done] read as priv->tx_bd_tail at
this point ? The value set a few lines above or some older value ?

		/* Matches barrier in nixge_start_xmit_done */
		smp_mb();

		/* Space might have just been freed - check again */
		if (!nixge_check_tx_bd_space(priv, MAX_SKB_FRAGS + 1))
			netif_wake_queue(ndev);
	}

-- 
Ueimor