mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Ben Hutchings <bhutchings@solarflare.com>
To: Rob Herring <robherring2@gmail.com>
Cc: <netdev@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
	Andreas Herrmann <andreas.herrmann@calxeda.com>,
	Lennert Buytenhek <buytenh@wantstofly.org>,
	Rob Herring <rob.herring@calxeda.com>
Subject: Re: [PATCH 06/11] net: calxedaxgmac: fix race with tx queue stop/wake
Date: Tue, 27 Aug 2013 19:31:57 +0100	[thread overview]
Message-ID: <1377628317.13272.64.camel@bwh-desktop.uk.level5networks.com> (raw)
In-Reply-To: <1377557126-10716-6-git-send-email-robherring2@gmail.com>

On Mon, 2013-08-26 at 17:45 -0500, Rob Herring wrote:
> From: Rob Herring <rob.herring@calxeda.com>
> 
> Since the xgmac transmit start and completion work locklessly, it is
> possible for xgmac_xmit to stop the tx queue after the xgmac_tx_complete
> has run resulting in the tx queue never being woken up. Fix this by
> ensuring that ring buffer index updates are visible and serialize the
> queue wake with netif_tx_lock.
>
> The implementation used here was copied from
> drivers/net/ethernet/broadcom/tg3.c.
> 
> Signed-off-by: Rob Herring <rob.herring@calxeda.com>
> ---
>  drivers/net/ethernet/calxeda/xgmac.c | 25 +++++++++++++++++++------
>  1 file changed, 19 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/ethernet/calxeda/xgmac.c b/drivers/net/ethernet/calxeda/xgmac.c
> index f630855..cd5872c 100644
> --- a/drivers/net/ethernet/calxeda/xgmac.c
> +++ b/drivers/net/ethernet/calxeda/xgmac.c
> @@ -410,6 +410,9 @@ struct xgmac_priv {
>  #define dma_ring_space(h, t, s)	CIRC_SPACE(h, t, s)
>  #define dma_ring_cnt(h, t, s)	CIRC_CNT(h, t, s)
>  
> +#define tx_dma_ring_space(p) \
> +	dma_ring_space((p)->tx_head, (p)->tx_tail, DMA_TX_RING_SZ)
> +
>  /* XGMAC Descriptor Access Helpers */
>  static inline void desc_set_buf_len(struct xgmac_dma_desc *p, u32 buf_sz)
>  {
> @@ -886,9 +889,14 @@ static void xgmac_tx_complete(struct xgmac_priv *priv)
>  		priv->tx_tail = dma_ring_incr(entry, DMA_TX_RING_SZ);
>  	}
>  
> -	if (dma_ring_space(priv->tx_head, priv->tx_tail, DMA_TX_RING_SZ) >
> -	    MAX_SKB_FRAGS)
> -		netif_wake_queue(priv->dev);
> +	/* Ensure tx_tail is visible to xgmac_xmit */
> +	smp_mb();
> +	if (unlikely(netif_queue_stopped(priv->dev))) {
> +		netif_tx_lock(priv->dev);
> +		if (tx_dma_ring_space(priv) > MAX_SKB_FRAGS)
> +			netif_wake_queue(priv->dev);
> +		netif_tx_unlock(priv->dev);
> +	}
>  }

You don't need to take the TX lock for this.  The memory barriers
provide sufficient synchronisation.

>  static void xgmac_tx_timeout_work(struct work_struct *work)
> @@ -1125,10 +1133,15 @@ static netdev_tx_t xgmac_xmit(struct sk_buff *skb, struct net_device *dev)
>  
>  	priv->tx_head = dma_ring_incr(entry, DMA_TX_RING_SZ);
>  
> -	if (dma_ring_space(priv->tx_head, priv->tx_tail, DMA_TX_RING_SZ) <
> -	    MAX_SKB_FRAGS)
> +	/* Ensure tx_head update is visible to tx completion */
> +	smp_mb();
> +	if (unlikely(tx_dma_ring_space(priv) < MAX_SKB_FRAGS)) {
>  		netif_stop_queue(dev);
> -
> +		/* Ensure netif_stop_queue is visible to tx completion */
> +		smp_mb();
> +		if (tx_dma_ring_space(priv) > MAX_SKB_FRAGS)
> +			netif_wake_queue(dev);

You should use netif_start_queue() rather than netif_wake_queue(), since
you know the TX scheduler is already active.

Ben.

> +	}
>  	return NETDEV_TX_OK;
>  }
>  

-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.


  reply	other threads:[~2013-08-27 18:32 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-08-26 22:45 [PATCH 01/11] net: calxedaxgmac: remove NETIF_F_FRAGLIST setting Rob Herring
2013-08-26 22:45 ` [PATCH 02/11] net: calxedaxgmac: read correct field in xgmac_desc_get_buf_len Rob Herring
2013-08-26 22:45 ` [PATCH 03/11] net: calxedaxgmac: fix race between xgmac_tx_complete and xgmac_tx_err Rob Herring
2013-08-26 22:45 ` [PATCH 04/11] net: calxedaxgmac: fix possible skb free before tx complete Rob Herring
2013-08-26 22:45 ` [PATCH 05/11] net: calxedaxgmac: update ring buffer tx_head after barriers Rob Herring
2013-08-26 22:45 ` [PATCH 06/11] net: calxedaxgmac: fix race with tx queue stop/wake Rob Herring
2013-08-27 18:31   ` Ben Hutchings [this message]
2013-08-26 22:45 ` [PATCH 07/11] net: calxedaxgmac: enable interrupts after napi_enable Rob Herring
2013-08-26 22:45 ` [PATCH 08/11] net: calxedaxgmac: fix various errors in xgmac_set_rx_mode Rob Herring
2013-08-26 22:45 ` [PATCH 09/11] net: calxedaxgmac: remove some unused statistic counters Rob Herring
2013-08-26 22:45 ` [PATCH 10/11] net: calxedaxgmac: fix rx DMA mapping API size mismatches Rob Herring
2013-08-26 22:45 ` [PATCH 11/11] net: calxedaxgmac: fix xgmac_xmit DMA mapping error handling Rob Herring

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1377628317.13272.64.camel@bwh-desktop.uk.level5networks.com \
    --to=bhutchings@solarflare.com \
    --cc=andreas.herrmann@calxeda.com \
    --cc=buytenh@wantstofly.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=netdev@vger.kernel.org \
    --cc=rob.herring@calxeda.com \
    --cc=robherring2@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®