From owner-svn-src-head@FreeBSD.ORG Wed Oct 31 23:50:36 2012 Return-Path: Delivered-To: svn-src-head@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [69.147.83.52]) by hub.freebsd.org (Postfix) with ESMTP id C3F48957; Wed, 31 Oct 2012 23:50:36 +0000 (UTC) (envelope-from jfv@FreeBSD.org) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:4f8:fff6::2c]) by mx1.freebsd.org (Postfix) with ESMTP id AACDA8FC12; Wed, 31 Oct 2012 23:50:36 +0000 (UTC) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.4/8.14.4) with ESMTP id q9VNoa01065185; Wed, 31 Oct 2012 23:50:36 GMT (envelope-from jfv@svn.freebsd.org) Received: (from jfv@localhost) by svn.freebsd.org (8.14.4/8.14.4/Submit) id q9VNoaAa065183; Wed, 31 Oct 2012 23:50:36 GMT (envelope-from jfv@svn.freebsd.org) Message-Id: <201210312350.q9VNoaAa065183@svn.freebsd.org> From: Jack F Vogel Date: Wed, 31 Oct 2012 23:50:36 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r242421 - head/sys/dev/ixgbe X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-head@freebsd.org X-Mailman-Version: 2.1.14 Precedence: list List-Id: SVN commit messages for the src tree for head/-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 31 Oct 2012 23:50:36 -0000 Author: jfv Date: Wed Oct 31 23:50:36 2012 New Revision: 242421 URL: http://svn.freebsd.org/changeset/base/242421 Log: A few important fixes: - Testing TSO6 has led me to discover that HW RSC is a problematic feature, it is ONLY designed to work with IPv4 in the first place, and if IP forwarding is done it can't be disabled as LRO in the stack, also initial testing we've done at Intel shows an equal performance using TSO[46] on the TX and LRO on RX, if you ran older code on 82599 or later hardware you actually could have detrimental performance for this reason. So I am disabling the feature by default and all our adapters will now use LRO instead. - If you have flow control off and multiple queues it was possible when the buffer of one queue becomes full that all RX movement is stalled, to eliminate this problem a feature bit is now set that will allow packets to be dropped when full rather than stall. Note, the default is to have flow control on, and this keeps this from happening. - Because of the recent fixes in the stack, LRO is now auto-disabled when problematic, so I have decided to enable it by default in the capabilities in the driver. - There are some 1G modules used by some customers, a couple small tweaks to properly support those in the media code. - A note: we have now done some testing of TSO6 and using LRO with IPv6 and it all works great!! Seeing line rate in both directions in best cases. Thanks bz for your excellent work!! Modified: head/sys/dev/ixgbe/ixgbe.c Modified: head/sys/dev/ixgbe/ixgbe.c ============================================================================== --- head/sys/dev/ixgbe/ixgbe.c Wed Oct 31 23:44:19 2012 (r242420) +++ head/sys/dev/ixgbe/ixgbe.c Wed Oct 31 23:50:36 2012 (r242421) @@ -47,7 +47,7 @@ int ixgbe_display_debug_stat /********************************************************************* * Driver version *********************************************************************/ -char ixgbe_driver_version[] = "2.4.8"; +char ixgbe_driver_version[] = "2.4.11"; /********************************************************************* * PCI Device ID Table @@ -181,6 +181,9 @@ static __inline void ixgbe_rx_discard(st static __inline void ixgbe_rx_input(struct rx_ring *, struct ifnet *, struct mbuf *, u32); +static void ixgbe_enable_rx_drop(struct adapter *); +static void ixgbe_disable_rx_drop(struct adapter *); + /* Support for pluggable optic modules */ static bool ixgbe_sfp_probe(struct adapter *); static void ixgbe_setup_optics(struct adapter *); @@ -292,6 +295,20 @@ TUNABLE_INT("hw.ixgbe.txd", &ixgbe_txd); static int ixgbe_rxd = PERFORM_RXD; TUNABLE_INT("hw.ixgbe.rxd", &ixgbe_rxd); +/* +** HW RSC control: +** this feature only works with +** IPv4, and only on 82599 and later. +** Also this will cause IP forwarding to +** fail and that can't be controlled by +** the stack as LRO can. For all these +** reasons I've deemed it best to leave +** this off and not bother with a tuneable +** interface, this would need to be compiled +** to enable. +*/ +static bool ixgbe_rsc_enable = FALSE; + /* Keep running tab on them for sanity check */ static int ixgbe_total_ports; @@ -1667,7 +1684,7 @@ ixgbe_media_status(struct ifnet * ifp, s ifmr->ifm_active |= IFM_100_TX | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: - ifmr->ifm_active |= IFM_1000_T | IFM_FDX; + ifmr->ifm_active |= adapter->optics | IFM_FDX; break; case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= adapter->optics | IFM_FDX; @@ -2035,7 +2052,6 @@ ixgbe_local_timer(void *arg) /* ** Check the TX queues status - ** - central locked handling of OACTIVE ** - watchdog only if all queues show hung */ for (int i = 0; i < adapter->num_queues; i++, que++, txr++) { @@ -2210,6 +2226,11 @@ ixgbe_setup_optics(struct adapter *adapt return; } + if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_SX) { + adapter->optics = IFM_1000_SX; + return; + } + if (layer & (IXGBE_PHYSICAL_LAYER_10GBASE_LR | IXGBE_PHYSICAL_LAYER_10GBASE_LRM)) { adapter->optics = IFM_10G_LR; @@ -2617,14 +2638,12 @@ ixgbe_setup_interface(device_t dev, stru ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_TSO | IFCAP_VLAN_HWCSUM; ifp->if_capabilities |= IFCAP_JUMBO_MTU; + ifp->if_capabilities |= IFCAP_LRO; ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWTSO | IFCAP_VLAN_MTU; ifp->if_capenable = ifp->if_capabilities; - /* Don't enable LRO by default */ - ifp->if_capabilities |= IFCAP_LRO; - /* ** Don't turn this on by default, if vlans are ** created on another pseudo device (eg. lagg) @@ -3899,6 +3918,10 @@ ixgbe_rsc_count(union ixgbe_adv_rx_desc * for an RX ring, this is toggled by the LRO capability * even though it is transparent to the stack. * + * NOTE: since this HW feature only works with IPV4 and + * our testing has shown soft LRO to be as effective + * I have decided to disable this by default. + * **********************************************************************/ static void ixgbe_setup_hw_rsc(struct rx_ring *rxr) @@ -3907,6 +3930,13 @@ ixgbe_setup_hw_rsc(struct rx_ring *rxr) struct ixgbe_hw *hw = &adapter->hw; u32 rscctrl, rdrxctl; + /* If turning LRO/RSC off we need to disable it */ + if ((adapter->ifp->if_capenable & IFCAP_LRO) == 0) { + rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxr->me)); + rscctrl &= ~IXGBE_RSCCTL_RSCEN; + return; + } + rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL); rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE; #ifdef DEV_NETMAP /* crcstrip is optional in netmap */ @@ -4108,12 +4138,8 @@ skip_head: /* ** Now set up the LRO interface: - ** 82598 uses software LRO, the - ** 82599 and X540 use a hardware assist. */ - if ((adapter->hw.mac.type != ixgbe_mac_82598EB) && - (ifp->if_capenable & IFCAP_RXCSUM) && - (ifp->if_capenable & IFCAP_LRO)) + if (ixgbe_rsc_enable) ixgbe_setup_hw_rsc(rxr); else if (ifp->if_capenable & IFCAP_LRO) { int err = tcp_lro_init(lro); @@ -5737,10 +5763,14 @@ ixgbe_set_flowcntl(SYSCTL_HANDLER_ARGS) case ixgbe_fc_tx_pause: case ixgbe_fc_full: adapter->hw.fc.requested_mode = adapter->fc; + if (adapter->num_queues > 1) + ixgbe_disable_rx_drop(adapter); break; case ixgbe_fc_none: default: adapter->hw.fc.requested_mode = ixgbe_fc_none; + if (adapter->num_queues > 1) + ixgbe_enable_rx_drop(adapter); } /* Don't autoneg if forcing a value */ adapter->hw.fc.disable_fc_autoneg = TRUE; @@ -5837,3 +5867,34 @@ ixgbe_set_thermal_test(SYSCTL_HANDLER_AR return (0); } + +/* +** Enable the hardware to drop packets when the buffer is +** full. This is useful when multiqueue,so that no single +** queue being full stalls the entire RX engine. We only +** enable this when Multiqueue AND when Flow Control is +** disabled. +*/ +static void +ixgbe_enable_rx_drop(struct adapter *adapter) +{ + struct ixgbe_hw *hw = &adapter->hw; + + for (int i = 0; i < adapter->num_queues; i++) { + u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i)); + srrctl |= IXGBE_SRRCTL_DROP_EN; + IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(i), srrctl); + } +} + +static void +ixgbe_disable_rx_drop(struct adapter *adapter) +{ + struct ixgbe_hw *hw = &adapter->hw; + + for (int i = 0; i < adapter->num_queues; i++) { + u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i)); + srrctl &= ~IXGBE_SRRCTL_DROP_EN; + IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(i), srrctl); + } +}