From mboxrd@z Thu Jan 1 00:00:00 1970 From: Luis Henriques Subject: Re: [net PATCH] atl1c: Fix misuse of netdev_alloc_skb in refilling rx ring Date: Tue, 30 Jul 2013 09:53:02 +0100 Message-ID: <87fvuwh1jl.fsf@canonical.com> References: <87k3kbdcmy.fsf@canonical.com> <1374960610.3607.13.camel@deadeye.wl.decadent.org.uk> <1374969583.3669.23.camel@edumazet-glaptop> <20130727.200205.67471633133830510.davem@davemloft.net> <20130728104446.GB9876@neilslaptop.think-freely.org> <1375028154.3669.30.camel@edumazet-glaptop> <20130728185318.GA10795@neilslaptop.think-freely.org> <1375042972.2546.17.camel@deadeye.wl.decadent.org.uk> <1375052482.3669.54.camel@edumazet-glaptop> <1375053654.3669.58.camel@edumazet-glaptop> <877gg9wv08.fsf@canonical.com> <1375095442.3669.67.camel@edumazet-glaptop> <87y58pva88.fsf@canonical.com> <1375111804.10515.1.camel@edumazet-glaptop> <1375118644.10515.18.camel@edumazet-glaptop> Mime-Version: 1.0 Content-Type: text/plain Cc: Ben Hutchings , Neil Horman , David Miller , netdev@vger.kernel.org, jcliburn@gmail.com To: Eric Dumazet Return-path: Received: from youngberry.canonical.com ([91.189.89.112]:52679 "EHLO youngberry.canonical.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755432Ab3G3IxJ (ORCPT ); Tue, 30 Jul 2013 04:53:09 -0400 In-Reply-To: <1375118644.10515.18.camel@edumazet-glaptop> (Eric Dumazet's message of "Mon, 29 Jul 2013 10:24:04 -0700") Sender: netdev-owner@vger.kernel.org List-ID: Eric Dumazet writes: > From: Eric Dumazet > > On Mon, 2013-07-29 at 08:30 -0700, Eric Dumazet wrote: >> On Mon, 2013-07-29 at 13:09 +0100, Luis Henriques wrote: >> >> > >> > I confirm that I can't reproduce the issue using this patch. >> > >> >> Thanks, I'll send a polished patch, as this one had an error if >> build_skb() returns NULL (in case sk_buff allocation fails) > > Please try the following patch : It should use 2K frags instead of 4K > for normal 1500 mtu This patch seems to work fine as well -- I'm unable to reproduce the issue. Cheers, -- Luis > > Thanks ! > > [PATCH] atl1c: use custom skb allocator > > We had reports ( https://bugzilla.kernel.org/show_bug.cgi?id=54021 ) > that using high order pages for skb allocations is problematic for atl1c > > We do not know exactly what the problem is, but we suspect that crossing > 4K pages is not well supported by this hardware. > > Use a custom allocator, using page allocator and 2K fragments for > optimal stack behavior. We might make this allocator generic > in future kernels. > > Signed-off-by: Eric Dumazet > Cc: Luis Henriques > Cc: Neil Horman > --- > drivers/net/ethernet/atheros/atl1c/atl1c.h | 3 + > drivers/net/ethernet/atheros/atl1c/atl1c_main.c | 40 +++++++++++++- > 2 files changed, 42 insertions(+), 1 deletion(-) > > diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c.h b/drivers/net/ethernet/atheros/atl1c/atl1c.h > index b2bf324..0f05565 100644 > --- a/drivers/net/ethernet/atheros/atl1c/atl1c.h > +++ b/drivers/net/ethernet/atheros/atl1c/atl1c.h > @@ -520,6 +520,9 @@ struct atl1c_adapter { > struct net_device *netdev; > struct pci_dev *pdev; > struct napi_struct napi; > + struct page *rx_page; > + unsigned int rx_page_offset; > + unsigned int rx_frag_size; > struct atl1c_hw hw; > struct atl1c_hw_stats hw_stats; > struct mii_if_info mii; /* MII interface info */ > diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c > index 786a874..a36a760 100644 > --- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c > +++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c > @@ -481,10 +481,15 @@ static int atl1c_set_mac_addr(struct net_device *netdev, void *p) > static void atl1c_set_rxbufsize(struct atl1c_adapter *adapter, > struct net_device *dev) > { > + unsigned int head_size; > int mtu = dev->mtu; > > adapter->rx_buffer_len = mtu > AT_RX_BUF_SIZE ? > roundup(mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN, 8) : AT_RX_BUF_SIZE; > + > + head_size = SKB_DATA_ALIGN(adapter->rx_buffer_len + NET_SKB_PAD) + > + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); > + adapter->rx_frag_size = roundup_pow_of_two(head_size); > } > > static netdev_features_t atl1c_fix_features(struct net_device *netdev, > @@ -952,6 +957,10 @@ static void atl1c_free_ring_resources(struct atl1c_adapter *adapter) > kfree(adapter->tpd_ring[0].buffer_info); > adapter->tpd_ring[0].buffer_info = NULL; > } > + if (adapter->rx_page) { > + put_page(adapter->rx_page); > + adapter->rx_page = NULL; > + } > } > > /** > @@ -1639,6 +1648,35 @@ static inline void atl1c_rx_checksum(struct atl1c_adapter *adapter, > skb_checksum_none_assert(skb); > } > > +static struct sk_buff *atl1c_alloc_skb(struct atl1c_adapter *adapter) > +{ > + struct sk_buff *skb; > + struct page *page; > + > + if (adapter->rx_frag_size > PAGE_SIZE) > + return netdev_alloc_skb(adapter->netdev, > + adapter->rx_buffer_len); > + > + page = adapter->rx_page; > + if (!page) { > + adapter->rx_page = page = alloc_page(GFP_ATOMIC); > + if (unlikely(!page)) > + return NULL; > + adapter->rx_page_offset = 0; > + } > + > + skb = build_skb(page_address(page) + adapter->rx_page_offset, > + adapter->rx_frag_size); > + if (likely(skb)) { > + adapter->rx_page_offset += adapter->rx_frag_size; > + if (adapter->rx_page_offset >= PAGE_SIZE) > + adapter->rx_page = NULL; > + else > + get_page(page); > + } > + return skb; > +} > + > static int atl1c_alloc_rx_buffer(struct atl1c_adapter *adapter) > { > struct atl1c_rfd_ring *rfd_ring = &adapter->rfd_ring; > @@ -1660,7 +1698,7 @@ static int atl1c_alloc_rx_buffer(struct atl1c_adapter *adapter) > while (next_info->flags & ATL1C_BUFFER_FREE) { > rfd_desc = ATL1C_RFD_DESC(rfd_ring, rfd_next_to_use); > > - skb = netdev_alloc_skb(adapter->netdev, adapter->rx_buffer_len); > + skb = atl1c_alloc_skb(adapter); > if (unlikely(!skb)) { > if (netif_msg_rx_err(adapter)) > dev_warn(&pdev->dev, "alloc rx buffer failed\n"); >