From owner-svn-src-all@FreeBSD.ORG Tue Nov 23 22:12:02 2010 Return-Path: Delivered-To: svn-src-all@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 8AC9C106567A; Tue, 23 Nov 2010 22:12:02 +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 78A1F8FC12; Tue, 23 Nov 2010 22:12:02 +0000 (UTC) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.3/8.14.3) with ESMTP id oANMC2Le045868; Tue, 23 Nov 2010 22:12:02 GMT (envelope-from jfv@svn.freebsd.org) Received: (from jfv@localhost) by svn.freebsd.org (8.14.3/8.14.3/Submit) id oANMC2EK045865; Tue, 23 Nov 2010 22:12:02 GMT (envelope-from jfv@svn.freebsd.org) Message-Id: <201011232212.oANMC2EK045865@svn.freebsd.org> From: Jack F Vogel Date: Tue, 23 Nov 2010 22:12:02 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cc: Subject: svn commit: r215781 - head/sys/dev/e1000 X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 23 Nov 2010 22:12:02 -0000 Author: jfv Date: Tue Nov 23 22:12:02 2010 New Revision: 215781 URL: http://svn.freebsd.org/changeset/base/215781 Log: - New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2 Modified: head/sys/dev/e1000/if_igb.c head/sys/dev/e1000/if_igb.h Modified: head/sys/dev/e1000/if_igb.c ============================================================================== --- head/sys/dev/e1000/if_igb.c Tue Nov 23 22:07:10 2010 (r215780) +++ head/sys/dev/e1000/if_igb.c Tue Nov 23 22:12:02 2010 (r215781) @@ -99,7 +99,7 @@ int igb_display_debug_stats = 0; /********************************************************************* * Driver version: *********************************************************************/ -char igb_driver_version[] = "version - 2.0.4"; +char igb_driver_version[] = "version - 2.0.7"; /********************************************************************* @@ -137,6 +137,10 @@ static igb_vendor_info_t igb_vendor_info { 0x8086, E1000_DEV_ID_82580_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82580_COPPER_DUAL, PCI_ANY_ID, PCI_ANY_ID, 0}, + { 0x8086, E1000_DEV_ID_82580_QUAD_FIBER, + PCI_ANY_ID, PCI_ANY_ID, 0}, + { 0x8086, E1000_DEV_ID_DH89XXCC_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0}, + { 0x8086, E1000_DEV_ID_DH89XXCC_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0}, /* required last entry */ { 0, 0, 0, 0, 0} }; @@ -307,11 +311,19 @@ static int igb_enable_msix = 1; TUNABLE_INT("hw.igb.enable_msix", &igb_enable_msix); /* - * Header split has seemed to be beneficial in - * many circumstances tested, however there have - * been some stability issues, so the default is - * off. - */ +** Tuneable Interrupt rate +*/ +static int igb_max_interrupt_rate = 8000; +TUNABLE_INT("hw.igb.max_interrupt_rate", &igb_max_interrupt_rate); + +/* +** Header split causes the packet header to +** be dma'd to a seperate mbuf from the payload. +** this can have memory alignment benefits. But +** another plus is that small packets often fit +** into the header and thus use no cluster. Its +** a very workload dependent type feature. +*/ static bool igb_header_split = FALSE; TUNABLE_INT("hw.igb.hdr_split", &igb_header_split); @@ -330,15 +342,6 @@ TUNABLE_INT("hw.igb.rx_process_limit", & static int igb_fc_setting = e1000_fc_full; TUNABLE_INT("hw.igb.fc_setting", &igb_fc_setting); -/* -** Shadow VFTA table, this is needed because -** the real filter table gets cleared during -** a soft reset and the driver needs to be able -** to repopulate it. -*/ -static u32 igb_shadow_vfta[IGB_VFTA_SIZE]; - - /********************************************************************* * Device identification routine * @@ -818,7 +821,7 @@ igb_start_locked(struct tx_ring *txr, st /* Set watchdog on */ txr->watchdog_time = ticks; - txr->watchdog_check = TRUE; + txr->queue_status = IGB_QUEUE_WORKING; } } @@ -922,7 +925,7 @@ igb_mq_start_locked(struct ifnet *ifp, s } if (enq > 0) { /* Set the watchdog */ - txr->watchdog_check = TRUE; + txr->queue_status = IGB_QUEUE_WORKING; txr->watchdog_time = ticks; } return (err); @@ -1049,6 +1052,11 @@ igb_ioctl(struct ifnet *ifp, u_long comm } break; case SIOCSIFMEDIA: + /* + ** As the speed/duplex settings are being + ** changed, we need toreset the PHY. + */ + adapter->hw.phy.reset_disable = FALSE; /* Check SOL/IDER usage */ IGB_CORE_LOCK(adapter); if (e1000_check_reset_block(&adapter->hw)) { @@ -1161,19 +1169,6 @@ igb_init_locked(struct adapter *adapter) E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN); - /* Use real VLAN Filter support? */ - if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { - if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) - /* Use real VLAN Filter support */ - igb_setup_vlan_hw_support(adapter); - else { - u32 ctrl; - ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); - ctrl |= E1000_CTRL_VME; - E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); - } - } - /* Set hardware offload abilities */ ifp->if_hwassist = 0; if (ifp->if_capenable & IFCAP_TXCSUM) { @@ -1201,10 +1196,12 @@ igb_init_locked(struct adapter *adapter) ** Figure out the desired mbuf pool ** for doing jumbo/packetsplit */ - if (ifp->if_mtu > ETHERMTU) + if (adapter->max_frame_size <= 2048) + adapter->rx_mbuf_sz = MCLBYTES; + else if (adapter->max_frame_size <= 4096) adapter->rx_mbuf_sz = MJUMPAGESIZE; else - adapter->rx_mbuf_sz = MCLBYTES; + adapter->rx_mbuf_sz = MJUM9BYTES; /* Prepare receive descriptors and buffers */ if (igb_setup_receive_structures(adapter)) { @@ -1213,6 +1210,19 @@ igb_init_locked(struct adapter *adapter) } igb_initialize_receive_units(adapter); + /* Use real VLAN Filter support? */ + if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { + if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) + /* Use real VLAN Filter support */ + igb_setup_vlan_hw_support(adapter); + else { + u32 ctrl; + ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); + ctrl |= E1000_CTRL_VME; + E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); + } + } + /* Don't lose promiscuous settings */ igb_set_promisc(adapter); @@ -1225,9 +1235,6 @@ igb_init_locked(struct adapter *adapter) if (adapter->msix > 1) /* Set up queue routing */ igb_configure_queues(adapter); - /* Set up VLAN tag offload and filter */ - igb_setup_vlan_hw_support(adapter); - /* this clears any pending interrupts */ E1000_READ_REG(&adapter->hw, E1000_ICR); #ifdef DEVICE_POLLING @@ -1627,11 +1634,6 @@ igb_media_change(struct ifnet *ifp) device_printf(adapter->dev, "Unsupported media type\n"); } - /* As the speed/duplex settings my have changed we need to - * reset the PHY. - */ - adapter->hw.phy.reset_disable = FALSE; - igb_init_locked(adapter); IGB_CORE_UNLOCK(adapter); @@ -1947,18 +1949,9 @@ igb_local_timer(void *arg) /* ** Watchdog: check for time since any descriptor was cleaned */ - for (int i = 0; i < adapter->num_queues; i++, txr++) { - IGB_TX_LOCK(txr); - if ((txr->watchdog_check == FALSE) || - (txr->tx_avail == adapter->num_tx_desc)) { - IGB_TX_UNLOCK(txr); - continue; - } - if ((ticks - txr->watchdog_time) > IGB_WATCHDOG) + for (int i = 0; i < adapter->num_queues; i++, txr++) + if (txr->queue_status == IGB_QUEUE_HUNG) goto timeout; - IGB_TX_UNLOCK(txr); - } - out: callout_reset(&adapter->timer, hz, igb_local_timer, adapter); return; @@ -1973,7 +1966,6 @@ timeout: txr->me, txr->tx_avail, txr->next_to_clean); adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING; adapter->watchdog_events++; - IGB_TX_UNLOCK(txr); igb_init_locked(adapter); } @@ -2037,7 +2029,7 @@ igb_update_link_status(struct adapter *a if_link_state_change(ifp, LINK_STATE_DOWN); /* Turn off watchdogs */ for (int i = 0; i < adapter->num_queues; i++, txr++) - txr->watchdog_check = FALSE; + txr->queue_status = IGB_QUEUE_IDLE; } } @@ -2069,7 +2061,7 @@ igb_stop(void *arg) /* Unarm watchdog timer. */ for (int i = 0; i < adapter->num_queues; i++, txr++) { IGB_TX_LOCK(txr); - txr->watchdog_check = FALSE; + txr->queue_status = IGB_QUEUE_IDLE; IGB_TX_UNLOCK(txr); } @@ -2280,8 +2272,7 @@ igb_configure_queues(struct adapter *ada { struct e1000_hw *hw = &adapter->hw; struct igb_queue *que; - u32 tmp, ivar = 0; - u32 newitr = IGB_DEFAULT_ITR; + u32 tmp, ivar = 0, newitr = 0; /* First turn on RSS capability */ if (adapter->hw.mac.type > e1000_82575) @@ -2398,6 +2389,9 @@ igb_configure_queues(struct adapter *ada } /* Set the starting interrupt rate */ + if (igb_max_interrupt_rate > 0) + newitr = (4000000 / igb_max_interrupt_rate) & 0x7FFC; + if (hw->mac.type == e1000_82575) newitr |= newitr << 16; else @@ -2509,6 +2503,8 @@ igb_setup_msix(struct adapter *adapter) /* Manual override */ if (igb_num_queues != 0) queues = igb_num_queues; + if (queues > 8) /* max queues */ + queues = 8; /* Can have max of 4 queues on 82575 */ if ((adapter->hw.mac.type == e1000_82575) && (queues > 4)) @@ -2636,7 +2632,7 @@ igb_reset(struct adapter *adapter) fc->send_xon = TRUE; /* Set Flow control, use the tunable location if sane */ - if ((igb_fc_setting >= 0) || (igb_fc_setting < 4)) + if ((igb_fc_setting >= 0) && (igb_fc_setting < 4)) fc->requested_mode = igb_fc_setting; else fc->requested_mode = e1000_fc_none; @@ -2728,10 +2724,11 @@ igb_setup_interface(device_t dev, struct ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM; ifp->if_capabilities |= IFCAP_TSO4; ifp->if_capabilities |= IFCAP_JUMBO_MTU; - if (igb_header_split) - ifp->if_capabilities |= IFCAP_LRO; - ifp->if_capenable = ifp->if_capabilities; + + /* Don't enable LRO by default */ + ifp->if_capabilities |= IFCAP_LRO; + #ifdef DEVICE_POLLING ifp->if_capabilities |= IFCAP_POLLING; #endif @@ -3172,7 +3169,7 @@ igb_initialize_transmit_units(struct ada E1000_READ_REG(hw, E1000_TDBAL(i)), E1000_READ_REG(hw, E1000_TDLEN(i))); - txr->watchdog_check = FALSE; + txr->queue_status = IGB_QUEUE_IDLE; txdctl |= IGB_TX_PTHRESH; txdctl |= IGB_TX_HTHRESH << 8; @@ -3184,14 +3181,14 @@ igb_initialize_transmit_units(struct ada if (adapter->hw.mac.type == e1000_vfadapt) return; + e1000_config_collision_dist(hw); + /* Program the Transmit Control Register */ tctl = E1000_READ_REG(hw, E1000_TCTL); tctl &= ~E1000_TCTL_CT; tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN | (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT)); - e1000_config_collision_dist(hw); - /* This write will effectively turn on the transmit unit. */ E1000_WRITE_REG(hw, E1000_TCTL, tctl); } @@ -3500,16 +3497,19 @@ static bool igb_txeof(struct tx_ring *txr) { struct adapter *adapter = txr->adapter; - int first, last, done; + int first, last, done, processed; struct igb_tx_buffer *tx_buffer; struct e1000_tx_desc *tx_desc, *eop_desc; struct ifnet *ifp = adapter->ifp; IGB_TX_LOCK_ASSERT(txr); - if (txr->tx_avail == adapter->num_tx_desc) + if (txr->tx_avail == adapter->num_tx_desc) { + txr->queue_status = IGB_QUEUE_IDLE; return FALSE; + } + processed = 0; first = txr->next_to_clean; tx_desc = &txr->tx_base[first]; tx_buffer = &txr->tx_buffers[first]; @@ -3536,6 +3536,7 @@ igb_txeof(struct tx_ring *txr) tx_desc->lower.data = 0; tx_desc->buffer_addr = 0; ++txr->tx_avail; + ++processed; if (tx_buffer->m_head) { txr->bytes += @@ -3575,6 +3576,15 @@ igb_txeof(struct tx_ring *txr) txr->next_to_clean = first; + /* + ** Watchdog calculation, we know there's + ** work outstanding or the first return + ** would have been taken, so none processed + ** for too long indicates a hang. + */ + if ((!processed) && ((ticks - txr->watchdog_time) > IGB_WATCHDOG)) + txr->queue_status = IGB_QUEUE_HUNG; + /* * If we have enough room, clear IFF_DRV_OACTIVE * to tell the stack that it is OK to send packets. @@ -3583,7 +3593,7 @@ igb_txeof(struct tx_ring *txr) ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; /* All clean, turn off the watchdog */ if (txr->tx_avail == adapter->num_tx_desc) { - txr->watchdog_check = FALSE; + txr->queue_status = IGB_QUEUE_IDLE; return (FALSE); } } @@ -3615,51 +3625,59 @@ igb_refresh_mbufs(struct rx_ring *rxr, i cleaned = -1; /* Signify no completions */ while (i != limit) { rxbuf = &rxr->rx_buffers[i]; - if ((rxbuf->m_head == NULL) && (rxr->hdr_split)) { + /* No hdr mbuf used with header split off */ + if (rxr->hdr_split == FALSE) + goto no_split; + if (rxbuf->m_head == NULL) { mh = m_gethdr(M_DONTWAIT, MT_DATA); if (mh == NULL) goto update; - mh->m_pkthdr.len = mh->m_len = MHLEN; - mh->m_len = MHLEN; - mh->m_flags |= M_PKTHDR; - m_adj(mh, ETHER_ALIGN); - /* Get the memory mapping */ - error = bus_dmamap_load_mbuf_sg(rxr->htag, - rxbuf->hmap, mh, hseg, &nsegs, BUS_DMA_NOWAIT); - if (error != 0) { - printf("GET BUF: dmamap load" - " failure - %d\n", error); - m_free(mh); - goto update; - } - rxbuf->m_head = mh; - bus_dmamap_sync(rxr->htag, rxbuf->hmap, - BUS_DMASYNC_PREREAD); - rxr->rx_base[i].read.hdr_addr = - htole64(hseg[0].ds_addr); - } + } else + mh = rxbuf->m_head; + mh->m_pkthdr.len = mh->m_len = MHLEN; + mh->m_len = MHLEN; + mh->m_flags |= M_PKTHDR; + /* Get the memory mapping */ + error = bus_dmamap_load_mbuf_sg(rxr->htag, + rxbuf->hmap, mh, hseg, &nsegs, BUS_DMA_NOWAIT); + if (error != 0) { + printf("Refresh mbufs: hdr dmamap load" + " failure - %d\n", error); + m_free(mh); + rxbuf->m_head = NULL; + goto update; + } + rxbuf->m_head = mh; + bus_dmamap_sync(rxr->htag, rxbuf->hmap, + BUS_DMASYNC_PREREAD); + rxr->rx_base[i].read.hdr_addr = + htole64(hseg[0].ds_addr); +no_split: if (rxbuf->m_pack == NULL) { mp = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, adapter->rx_mbuf_sz); if (mp == NULL) goto update; - mp->m_pkthdr.len = mp->m_len = adapter->rx_mbuf_sz; - /* Get the memory mapping */ - error = bus_dmamap_load_mbuf_sg(rxr->ptag, - rxbuf->pmap, mp, pseg, &nsegs, BUS_DMA_NOWAIT); - if (error != 0) { - printf("GET BUF: dmamap load" - " failure - %d\n", error); - m_free(mp); - goto update; - } - rxbuf->m_pack = mp; - bus_dmamap_sync(rxr->ptag, rxbuf->pmap, - BUS_DMASYNC_PREREAD); - rxr->rx_base[i].read.pkt_addr = - htole64(pseg[0].ds_addr); + } else + mp = rxbuf->m_pack; + + mp->m_pkthdr.len = mp->m_len = adapter->rx_mbuf_sz; + /* Get the memory mapping */ + error = bus_dmamap_load_mbuf_sg(rxr->ptag, + rxbuf->pmap, mp, pseg, &nsegs, BUS_DMA_NOWAIT); + if (error != 0) { + printf("Refresh mbufs: payload dmamap load" + " failure - %d\n", error); + m_free(mp); + rxbuf->m_pack = NULL; + goto update; } + rxbuf->m_pack = mp; + bus_dmamap_sync(rxr->ptag, rxbuf->pmap, + BUS_DMASYNC_PREREAD); + rxr->rx_base[i].read.pkt_addr = + htole64(pseg[0].ds_addr); cleaned = i; /* Calculate next index */ @@ -3722,9 +3740,9 @@ igb_allocate_receive_buffers(struct rx_r BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ - MJUMPAGESIZE, /* maxsize */ + MJUM9BYTES, /* maxsize */ 1, /* nsegments */ - MJUMPAGESIZE, /* maxsegsize */ + MJUM9BYTES, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ @@ -3826,7 +3844,7 @@ igb_setup_receive_ring(struct rx_ring *r rxr->hdr_split = TRUE; /* Now replenish the ring mbufs */ - for (int j = 0; j != adapter->num_rx_desc; ++j) { + for (int j = 0; j < adapter->num_rx_desc; ++j) { struct mbuf *mh, *mp; rxbuf = &rxr->rx_buffers[j]; @@ -3897,8 +3915,8 @@ skip_head: ** are undesireable in similar setups. */ if (ifp->if_capenable & IFCAP_LRO) { - int err = tcp_lro_init(lro); - if (err) { + error = tcp_lro_init(lro); + if (error) { device_printf(dev, "LRO Initialization failed!\n"); goto fail; } @@ -3940,7 +3958,9 @@ fail: */ for (int j = 0; j > i; ++j) { rxr = &adapter->rx_rings[i]; + IGB_RX_LOCK(rxr); igb_free_receive_ring(rxr); + IGB_RX_UNLOCK(rxr); } return (ENOBUFS); @@ -3983,9 +4003,13 @@ igb_initialize_receive_units(struct adap */ if (ifp->if_mtu > ETHERMTU) { rctl |= E1000_RCTL_LPE; - srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT; - rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX; - + if (adapter->rx_mbuf_sz == MJUMPAGESIZE) { + srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT; + rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX; + } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) { + srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT; + rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX; + } /* Set maximum packet len */ psize = adapter->max_frame_size; /* are we on a vlan? */ @@ -4194,11 +4218,11 @@ igb_free_receive_buffers(struct rx_ring static __inline void igb_rx_discard(struct rx_ring *rxr, int i) { - struct adapter *adapter = rxr->adapter; struct igb_rx_buf *rbuf; - struct mbuf *mh, *mp; rbuf = &rxr->rx_buffers[i]; + + /* Partially received? Free the chain */ if (rxr->fmp != NULL) { rxr->fmp->m_flags |= M_PKTHDR; m_freem(rxr->fmp); @@ -4206,19 +4230,23 @@ igb_rx_discard(struct rx_ring *rxr, int rxr->lmp = NULL; } - mh = rbuf->m_head; - mp = rbuf->m_pack; + /* + ** With advanced descriptors the writeback + ** clobbers the buffer addrs, so its easier + ** to just free the existing mbufs and take + ** the normal refresh path to get new buffers + ** and mapping. + */ + if (rbuf->m_head) { + m_free(rbuf->m_head); + rbuf->m_head = NULL; + } - /* Reuse loaded DMA map and just update mbuf chain */ - if (mh) { /* with no hdr split would be null */ - mh->m_len = MHLEN; - mh->m_flags |= M_PKTHDR; - mh->m_next = NULL; + if (rbuf->m_pack) { + m_free(rbuf->m_pack); + rbuf->m_pack = NULL; } - mp->m_len = mp->m_pkthdr.len = adapter->rx_mbuf_sz; - mp->m_data = mp->m_ext.ext_buf; - mp->m_next = NULL; return; } @@ -4333,10 +4361,9 @@ igb_rxeof(struct igb_queue *que, int cou E1000_RXDADV_HDRBUFLEN_SHIFT; if (hlen > IGB_HDR_BUF) hlen = IGB_HDR_BUF; - /* Handle the header mbuf */ mh = rxr->rx_buffers[i].m_head; mh->m_len = hlen; - /* clear buf info for refresh */ + /* clear buf pointer for refresh */ rxbuf->m_head = NULL; /* ** Get the payload length, this @@ -4347,7 +4374,7 @@ igb_rxeof(struct igb_queue *que, int cou mp = rxr->rx_buffers[i].m_pack; mp->m_len = plen; mh->m_next = mp; - /* clear buf info for refresh */ + /* clear buf pointer */ rxbuf->m_pack = NULL; rxr->rx_split_packets++; } @@ -4368,7 +4395,7 @@ igb_rxeof(struct igb_queue *que, int cou /* Initial frame - setup */ if (rxr->fmp == NULL) { mh->m_pkthdr.len = mh->m_len; - /* Store the first mbuf */ + /* Save the head of the chain */ rxr->fmp = mh; rxr->lmp = mh; if (mp != NULL) { @@ -4533,12 +4560,15 @@ igb_register_vlan(void *arg, struct ifne if ((vtag == 0) || (vtag > 4095)) /* Invalid */ return; + IGB_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; - igb_shadow_vfta[index] |= (1 << bit); + adapter->shadow_vfta[index] |= (1 << bit); ++adapter->num_vlans; /* Re-init to load the changes */ - igb_init(adapter); + if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) + igb_init_locked(adapter); + IGB_CORE_UNLOCK(adapter); } /* @@ -4557,12 +4587,15 @@ igb_unregister_vlan(void *arg, struct if if ((vtag == 0) || (vtag > 4095)) /* Invalid */ return; + IGB_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; - igb_shadow_vfta[index] &= ~(1 << bit); + adapter->shadow_vfta[index] &= ~(1 << bit); --adapter->num_vlans; /* Re-init to load the changes */ - igb_init(adapter); + if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) + igb_init_locked(adapter); + IGB_CORE_UNLOCK(adapter); } static void @@ -4585,12 +4618,13 @@ igb_setup_vlan_hw_support(struct adapter ** we need to repopulate it now. */ for (int i = 0; i < IGB_VFTA_SIZE; i++) - if (igb_shadow_vfta[i] != 0) { + if (adapter->shadow_vfta[i] != 0) { if (hw->mac.type == e1000_vfadapt) - e1000_vfta_set_vf(hw, igb_shadow_vfta[i], TRUE); + e1000_vfta_set_vf(hw, + adapter->shadow_vfta[i], TRUE); else E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, - i, igb_shadow_vfta[i]); + i, adapter->shadow_vfta[i]); } if (hw->mac.type == e1000_vfadapt) @@ -4994,6 +5028,28 @@ igb_sysctl_reg_handler(SYSCTL_HANDLER_AR } /* +** Tuneable interrupt rate handler +*/ +static int +igb_sysctl_interrupt_rate_handler(SYSCTL_HANDLER_ARGS) +{ + struct igb_queue *que = ((struct igb_queue *)oidp->oid_arg1); + int error; + u32 reg, usec, rate; + + reg = E1000_READ_REG(&que->adapter->hw, E1000_EITR(que->msix)); + usec = ((reg & 0x7FFC) >> 2); + if (usec > 0) + rate = 1000000 / usec; + else + rate = 0; + error = sysctl_handle_int(oidp, &rate, 0, req); + if (error || !req->newptr) + return error; + return 0; +} + +/* * Add sysctl variables, one per statistic, to the system. */ static void @@ -5065,6 +5121,12 @@ igb_add_hw_stats(struct adapter *adapter CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); + SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "interrupt_rate", + CTLFLAG_RD, &adapter->queues[i], + sizeof(&adapter->queues[i]), + igb_sysctl_interrupt_rate_handler, + "IU", "Interrupt Rate"); + SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head", CTLFLAG_RD, adapter, E1000_TDH(txr->me), igb_sysctl_reg_handler, "IU", Modified: head/sys/dev/e1000/if_igb.h ============================================================================== --- head/sys/dev/e1000/if_igb.h Tue Nov 23 22:07:10 2010 (r215780) +++ head/sys/dev/e1000/if_igb.h Tue Nov 23 22:12:02 2010 (r215781) @@ -190,6 +190,9 @@ #define IGB_TX_BUFFER_SIZE ((uint32_t) 1514) #define IGB_FC_PAUSE_TIME 0x0680 #define IGB_EEPROM_APME 0x400; +#define IGB_QUEUE_IDLE 0 +#define IGB_QUEUE_WORKING 1 +#define IGB_QUEUE_HUNG 2 /* * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be @@ -237,7 +240,7 @@ /* Define the starting Interrupt rate per Queue */ #define IGB_INTS_PER_SEC 8000 -#define IGB_DEFAULT_ITR 1000000000/(IGB_INTS_PER_SEC * 256) +#define IGB_DEFAULT_ITR ((1000000/IGB_INTS_PER_SEC) << 2) #define IGB_LINK_ITR 2000 @@ -300,7 +303,7 @@ struct tx_ring { u32 bytes; u32 packets; - bool watchdog_check; + int queue_status; int watchdog_time; int tdt; int tdh; @@ -384,7 +387,15 @@ struct adapter { int wol; int has_manage; - /* Info about the board itself */ + /* + ** Shadow VFTA table, this is needed because + ** the real vlan filter table gets cleared during + ** a soft reset and the driver needs to be able + ** to repopulate it. + */ + u32 shadow_vfta[IGB_VFTA_SIZE]; + + /* Info about the interface */ u8 link_active; u16 link_speed; u16 link_duplex;