From patchwork Fri Oct 16 12:01:37 2009 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Asier Llano Palacios X-Patchwork-Id: 36198 X-Patchwork-Delegate: grant.likely@secretlab.ca Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from bilbo.ozlabs.org (localhost [127.0.0.1]) by ozlabs.org (Postfix) with ESMTP id 52642B7E4E for ; Fri, 16 Oct 2009 23:00:57 +1100 (EST) Received: from ey-out-1920.google.com (ey-out-1920.google.com [74.125.78.149]) by ozlabs.org (Postfix) with ESMTP id 79ACBB7BBF for ; Fri, 16 Oct 2009 23:00:49 +1100 (EST) Received: by ey-out-1920.google.com with SMTP id 3so370385eyh.38 for ; Fri, 16 Oct 2009 05:00:46 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:received:received:subject:from:to:cc :content-type:date:message-id:mime-version:x-mailer :content-transfer-encoding; bh=2aU0ZuP+N8VVj5rE+P5oVOS6W04eRGRjwDinwyImJ/U=; b=PfiQ20xlYD1RJ9rfWithYKhUwWSZBhmucWCgH9zlCIJ43caTk/nZGGvfMUEsYM1TnG YwO8q+0Dl161NHaNwbSz474IdC+DJk+MI9a5Q4PWZc56SSij6AOqIESliRGHMcVG9ZR+ eZKshGQ4FxD91hNKKZ6XPAXpThkP4NOMLpx2g= DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=subject:from:to:cc:content-type:date:message-id:mime-version :x-mailer:content-transfer-encoding; b=xKhmg4TPa7mSL8IB9YeV46ADZHpeDYJ3KEVDWacbGbDSdlUG17p7IAMNvT8GXN80Ml jglPnVEzipbuYCuesuPIjdj3ZE1keaR2I0YHGGkH+YbN5/pd1/zrlY6Y76Uom9T5B5vQ /QMF+tFb+Qhw7m/6CXUuac8Ue5hGTiVmGDb8M= Received: by 10.211.131.40 with SMTP id i40mr802686ebn.99.1255694446097; Fri, 16 Oct 2009 05:00:46 -0700 (PDT) Received: from ?128.127.51.61? ([89.6.239.150]) by mx.google.com with ESMTPS id 10sm423472eyd.18.2009.10.16.05.00.43 (version=SSLv3 cipher=RC4-MD5); Fri, 16 Oct 2009 05:00:44 -0700 (PDT) Subject: [PATCH] Found skb leak in mpc5200 fec on rxfifo error From: Asier Llano Palacios To: linuxppc-dev@lists.ozlabs.org, Grant Likely Date: Fri, 16 Oct 2009 14:01:37 +0200 Message-Id: <1255694497.4683.163.camel@allano> Mime-Version: 1.0 X-Mailer: Evolution 2.26.3 Cc: a.arzuaga@ziv.es X-BeenThere: linuxppc-dev@lists.ozlabs.org X-Mailman-Version: 2.1.12 Precedence: list List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: linuxppc-dev-bounces+patchwork-incoming=ozlabs.org@lists.ozlabs.org Errors-To: linuxppc-dev-bounces+patchwork-incoming=ozlabs.org@lists.ozlabs.org (I'm sorry about my previous message with a legal notice, this time I resent it with another mail server that doesn't include the legal notice for me) Hi all, I finally found why in my mpc5200 based system (and in the lite5200) I had leaks of skbs when receiving ethernet traffic at 100 Mbit/s so that we generated an rxfifo error (to be able to get it I even reduced the speed of the CPU using the PLL jumpers). The problem was: The reception packets are handled in the interrupt handler: mpc52xx_fec_rx_interrupt. So this interrupt is being handled all the time because we are receiving packets continuously. The reception fifo errors are handled in the interrupt handler: mpc52xx_fec_interrupt. This interruption is being handled very often because we are receiving more packets that what we can handle, and that generates a fifo reception error. So it is very usual that the mpc52xx_fec_interrupt handler is executed at the middle of the mpc52xx_fec_rx_interrupt handler. And I think that it is not designed for that purpose. The mpc52xx_fec_rx_interrupt uses bcom_retrieve_buffer and bcom_submit_next_buffer. The mpc52xx_fec_interrupt calls to mpc52xx_fec_reset which calls to mpc52xx_fec_free_rx_buffers and mpc52xx_fec_alloc_rx_buffers, which at the end call bcom_retrieve_buffer and bcom_submit_next_buffer. Then we have two problems because of the same origin: - bcom_retrieve_buffer and bcom_submit_next_buffer doesn't seem to be atomic, so they cannot be called from one interrupt nested from another one being executing those functions. - Even if they were atomic, the mpc52xx_fec_rx_interrupt and the mpc52xx_fec_reset functions are not intended to be called at the same time. Patch I managed to do: I've managed to perform a patch that avoids the leak disabling the irqs, but I have a remaining problem: The last packet to be transmited is waiting for a packet to be received in order to egress (I don't know why but I detected it experimentaly). Anyone with more experience can tell me what's wrong in this patch? Thank you in advance, Asier Signed-off-by: Asier Llano --- drivers/net/fec_mpc52xx.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-) diff -urpN linux-2.6.31.2/drivers/net/fec_mpc52xx.c linux-2.6.31.2-fec_mpc52xx_skb_leak/drivers/net/fec_mpc52xx.c ================================================================ --- linux-2.6.31.2/drivers/net/fec_mpc52xx.c +++ linux-2.6.31.2-fec_mpc52xx_skb_leak/drivers/net/fec_mpc52xx.c @@ -85,13 +85,20 @@ MODULE_PARM_DESC(debug, "debugging messa static void mpc52xx_fec_tx_timeout(struct net_device *dev) { + struct mpc52xx_fec_priv *priv = netdev_priv(dev); + unsigned long flags; + dev_warn(&dev->dev, "transmit timed out\n"); + spin_lock_irqsave(&priv->lock, flags); + mpc52xx_fec_reset(dev); dev->stats.tx_errors++; netif_wake_queue(dev); + + spin_unlock_irqrestore(&priv->lock, flags); } static void mpc52xx_fec_set_paddr(struct net_device *dev, u8 *mac) @@ -359,8 +366,9 @@ static irqreturn_t mpc52xx_fec_tx_interr { struct net_device *dev = dev_id; struct mpc52xx_fec_priv *priv = netdev_priv(dev); + unsigned long flags; - spin_lock(&priv->lock); + spin_lock_irqsave(&priv->lock, flags); while (bcom_buffer_done(priv->tx_dmatsk)) { struct sk_buff *skb; @@ -375,7 +383,7 @@ static irqreturn_t mpc52xx_fec_tx_interr netif_wake_queue(dev); - spin_unlock(&priv->lock); + spin_unlock_irqrestore(&priv->lock, flags); return IRQ_HANDLED; } @@ -384,6 +392,9 @@ static irqreturn_t mpc52xx_fec_rx_interr { struct net_device *dev = dev_id; struct mpc52xx_fec_priv *priv = netdev_priv(dev); + unsigned long flags; + + spin_lock_irqsave(&priv->lock, flags); while (bcom_buffer_done(priv->rx_dmatsk)) { struct sk_buff *skb; @@ -445,6 +456,8 @@ static irqreturn_t mpc52xx_fec_rx_interr bcom_submit_next_buffer(priv->rx_dmatsk, skb); } + spin_unlock_irqrestore(&priv->lock, flags); + return IRQ_HANDLED; } @@ -454,6 +467,7 @@ static irqreturn_t mpc52xx_fec_interrupt struct mpc52xx_fec_priv *priv = netdev_priv(dev); struct mpc52xx_fec __iomem *fec = priv->fec; u32 ievent; + unsigned long flags; ievent = in_be32(&fec->ievent); @@ -471,9 +485,14 @@ static irqreturn_t mpc52xx_fec_interrupt if (net_ratelimit() && (ievent & FEC_IEVENT_XFIFO_ERROR)) dev_warn(&dev->dev, "FEC_IEVENT_XFIFO_ERROR\n"); + spin_lock_irqsave(&priv->lock, flags); + mpc52xx_fec_reset(dev); netif_wake_queue(dev); + + spin_unlock_irqrestore(&priv->lock, flags); + return IRQ_HANDLED; }