igbvf: fix mac type for 82576
[dpdk.git] / lib / librte_pmd_e1000 / igb_rxtx.c
index 87fe56f..7fe1780 100644 (file)
@@ -1,40 +1,38 @@
 /*-
  *   BSD LICENSE
  * 
- *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
  *   All rights reserved.
  * 
- *   Redistribution and use in source and binary forms, with or without 
- *   modification, are permitted provided that the following conditions 
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
  *   are met:
  * 
- *     * Redistributions of source code must retain the above copyright 
+ *     * Redistributions of source code must retain the above copyright
  *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright 
- *       notice, this list of conditions and the following disclaimer in 
- *       the documentation and/or other materials provided with the 
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
  *       distribution.
- *     * Neither the name of Intel Corporation nor the names of its 
- *       contributors may be used to endorse or promote products derived 
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
  *       from this software without specific prior written permission.
  * 
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- * 
  */
 
 #include <sys/queue.h>
 
-#include <endian.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -196,6 +194,11 @@ struct igb_tx_queue {
 #define rte_packet_prefetch(p) do {} while(0)
 #endif
 
+/*
+ * Macro for VMDq feature for 1 GbE NIC.
+ */
+#define E1000_VMOLR_SIZE                       (8)
+
 /*********************************************************************
  *
  *  TX function
@@ -1082,8 +1085,13 @@ ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
        if (mz)
                return mz;
 
+#ifdef RTE_LIBRTE_XEN_DOM0
+       return rte_memzone_reserve_bounded(z_name, ring_size,
+                       socket_id, 0, IGB_ALIGN, RTE_PGSIZE_2M);
+#else
        return rte_memzone_reserve_aligned(z_name, ring_size,
                        socket_id, 0, IGB_ALIGN);
+#endif
 }
 
 static void
@@ -1130,16 +1138,16 @@ igb_reset_tx_queue_stat(struct igb_tx_queue *txq)
 static void
 igb_reset_tx_queue(struct igb_tx_queue *txq, struct rte_eth_dev *dev)
 {
+       static const union e1000_adv_tx_desc zeroed_desc = { .read = {
+                       .buffer_addr = 0}};
        struct igb_tx_entry *txe = txq->sw_ring;
-       uint32_t size;
        uint16_t i, prev;
        struct e1000_hw *hw;
 
        hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-       size = sizeof(union e1000_adv_tx_desc) * txq->nb_tx_desc;
        /* Zero out HW ring memory */
-       for (i = 0; i < size; i++) {
-               ((volatile char *)txq->tx_ring)[i] = 0;
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               txq->tx_ring[i] = zeroed_desc;
        }
 
        /* Initialize ring entries */
@@ -1230,15 +1238,20 @@ eth_igb_tx_queue_setup(struct rte_eth_dev *dev,
        txq->pthresh = tx_conf->tx_thresh.pthresh;
        txq->hthresh = tx_conf->tx_thresh.hthresh;
        txq->wthresh = tx_conf->tx_thresh.wthresh;
+       if (txq->wthresh > 0 && hw->mac.type == e1000_82576)
+               txq->wthresh = 1;
        txq->queue_id = queue_idx;
        txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
                queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
        txq->port_id = dev->data->port_id;
 
        txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(txq->reg_idx));
+#ifndef RTE_LIBRTE_XEN_DOM0
        txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
-       txq->tx_ring = (union e1000_adv_tx_desc *) tz->addr;
-
+#else
+       txq->tx_ring_phys_addr = rte_mem_phy2mch(tz->memseg_id, tz->phys_addr);
+#endif
+        txq->tx_ring = (union e1000_adv_tx_desc *) tz->addr;
        /* Allocate software ring */
        txq->sw_ring = rte_zmalloc("txq->sw_ring",
                                   sizeof(struct igb_tx_entry) * nb_desc,
@@ -1291,13 +1304,13 @@ eth_igb_rx_queue_release(void *rxq)
 static void
 igb_reset_rx_queue(struct igb_rx_queue *rxq)
 {
-       unsigned size;
+       static const union e1000_adv_rx_desc zeroed_desc = { .read = {
+                       .pkt_addr = 0}};
        unsigned i;
 
        /* Zero out HW ring memory */
-       size = sizeof(union e1000_adv_rx_desc) * rxq->nb_rx_desc;
-       for (i = 0; i < size; i++) {
-               ((volatile char *)rxq->rx_ring)[i] = 0;
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               rxq->rx_ring[i] = zeroed_desc;
        }
 
        rxq->rx_tail = 0;
@@ -1346,6 +1359,8 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->pthresh = rx_conf->rx_thresh.pthresh;
        rxq->hthresh = rx_conf->rx_thresh.hthresh;
        rxq->wthresh = rx_conf->rx_thresh.wthresh;
+       if (rxq->wthresh > 0 && hw->mac.type == e1000_82576)
+               rxq->wthresh = 1;
        rxq->drop_en = rx_conf->rx_drop_en;
        rxq->rx_free_thresh = rx_conf->rx_free_thresh;
        rxq->queue_id = queue_idx;
@@ -1368,7 +1383,11 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
        }
        rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(rxq->reg_idx));
        rxq->rdh_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDH(rxq->reg_idx));
+#ifndef RTE_LIBRTE_XEN_DOM0
        rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
+#else
+       rxq->rx_ring_phys_addr = rte_mem_phy2mch(rz->memseg_id, rz->phys_addr); 
+#endif 
        rxq->rx_ring = (union e1000_adv_rx_desc *) rz->addr;
 
        /* Allocate software ring. */
@@ -1391,26 +1410,46 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
 uint32_t 
 eth_igb_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
+#define IGB_RXQ_SCAN_INTERVAL 4
+       volatile union e1000_adv_rx_desc *rxdp;
        struct igb_rx_queue *rxq;
-       uint32_t nb_pkts_available;
-       uint32_t rx_rdh;
-       uint32_t rx_id;
+       uint32_t desc = 0;
 
        if (rx_queue_id >= dev->data->nb_rx_queues) {
-               PMD_RX_LOG(DEBUG,"Invalid RX queue_id=%d\n", rx_queue_id);
+               PMD_RX_LOG(ERR, "Invalid RX queue id=%d\n", rx_queue_id);
                return 0;
        }
 
        rxq = dev->data->rx_queues[rx_queue_id];
-       rx_id = (uint16_t) ((rxq->rx_tail == 0) ? (rxq->nb_rx_desc - 1) :
-                                                       (rxq->rx_tail - 1));
-       rx_rdh = E1000_PCI_REG(rxq->rdh_reg_addr);
-       if (rx_rdh > rx_id) 
-               nb_pkts_available = rx_rdh - rx_id;
-       else
-               nb_pkts_available = rx_rdh - rx_id + rxq->nb_rx_desc;
-       
-       return (nb_pkts_available);
+       rxdp = &(rxq->rx_ring[rxq->rx_tail]);
+
+       while ((desc < rxq->nb_rx_desc) &&
+               (rxdp->wb.upper.status_error & E1000_RXD_STAT_DD)) {
+               desc += IGB_RXQ_SCAN_INTERVAL;
+               rxdp += IGB_RXQ_SCAN_INTERVAL;
+               if (rxq->rx_tail + desc >= rxq->nb_rx_desc)
+                       rxdp = &(rxq->rx_ring[rxq->rx_tail +
+                               desc - rxq->nb_rx_desc]);
+       }
+
+       return 0;
+}
+
+int
+eth_igb_rx_descriptor_done(void *rx_queue, uint16_t offset)
+{
+       volatile union e1000_adv_rx_desc *rxdp;
+       struct igb_rx_queue *rxq = rx_queue;
+       uint32_t desc;
+
+       if (unlikely(offset >= rxq->nb_rx_desc))
+               return 0;
+       desc = rxq->rx_tail + offset;
+       if (desc >= rxq->nb_rx_desc)
+               desc -= rxq->nb_rx_desc;
+
+       rxdp = &rxq->rx_ring[desc];
+       return !!(rxdp->wb.upper.status_error & E1000_RXD_STAT_DD);
 }
 
 void
@@ -1549,6 +1588,118 @@ igb_rss_configure(struct rte_eth_dev *dev)
        E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
 }
 
+/*
+ * Check if the mac type support VMDq or not.
+ * Return 1 if it supports, otherwise, return 0.
+ */
+static int
+igb_is_vmdq_supported(const struct rte_eth_dev *dev)
+{
+       const struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       
+       switch (hw->mac.type) { 
+       case e1000_82576: 
+       case e1000_82580: 
+       case e1000_i350: 
+               return 1;
+       case e1000_82540: 
+       case e1000_82541: 
+       case e1000_82542: 
+       case e1000_82543: 
+       case e1000_82544: 
+       case e1000_82545: 
+       case e1000_82546: 
+       case e1000_82547: 
+       case e1000_82571: 
+       case e1000_82572: 
+       case e1000_82573: 
+       case e1000_82574: 
+       case e1000_82583: 
+       case e1000_i210: 
+       case e1000_i211: 
+       default:
+               PMD_INIT_LOG(ERR, "Cannot support VMDq feature\n");
+               return 0;
+       }
+}
+
+static int
+igb_vmdq_rx_hw_configure(struct rte_eth_dev *dev)
+{
+       struct rte_eth_vmdq_rx_conf *cfg;
+       struct e1000_hw *hw;
+       uint32_t mrqc, vt_ctl, vmolr, rctl;
+       int i;
+       PMD_INIT_LOG(DEBUG, ">>");
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       cfg = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
+
+       /* Check if mac type can support VMDq, return value of 0 means NOT support */
+       if (igb_is_vmdq_supported(dev) == 0)
+               return -1;
+
+       igb_rss_disable(dev);
+       
+       /* RCTL: eanble VLAN filter */
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl |= E1000_RCTL_VFE;
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+
+       /* MRQC: enable vmdq */
+       mrqc = E1000_READ_REG(hw, E1000_MRQC);
+       mrqc |= E1000_MRQC_ENABLE_VMDQ; 
+       E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
+       /* VTCTL:  pool selection according to VLAN tag */
+       vt_ctl = E1000_READ_REG(hw, E1000_VT_CTL);
+       if (cfg->enable_default_pool) 
+               vt_ctl |= (cfg->default_pool << E1000_VT_CTL_DEFAULT_POOL_SHIFT);
+       vt_ctl |= E1000_VT_CTL_IGNORE_MAC;
+       E1000_WRITE_REG(hw, E1000_VT_CTL, vt_ctl);
+       
+       /* 
+        * VMOLR: set STRVLAN as 1 if IGMAC in VTCTL is set as 1
+        * Both 82576 and 82580 support it 
+        */
+       if (hw->mac.type != e1000_i350) {
+               for (i = 0; i < E1000_VMOLR_SIZE; i++) {
+                       vmolr = E1000_READ_REG(hw, E1000_VMOLR(i));
+                       vmolr |= E1000_VMOLR_STRVLAN;
+                       E1000_WRITE_REG(hw, E1000_VMOLR(i), vmolr);
+               }
+       }
+
+       /* VFTA - enable all vlan filters */
+       for (i = 0; i < IGB_VFTA_SIZE; i++) 
+               E1000_WRITE_REG(hw, (E1000_VFTA+(i*4)), UINT32_MAX);
+       
+       /* VFRE: 8 pools enabling for rx, both 82576 and i350 support it */
+       if (hw->mac.type != e1000_82580)
+               E1000_WRITE_REG(hw, E1000_VFRE, E1000_MBVFICR_VFREQ_MASK);
+       /*
+        * RAH/RAL - allow pools to read specific mac addresses
+        * In this case, all pools should be able to read from mac addr 0
+        */
+       E1000_WRITE_REG(hw, E1000_RAH(0), (E1000_RAH_AV | UINT16_MAX));
+       E1000_WRITE_REG(hw, E1000_RAL(0), UINT32_MAX);
+
+       /* VLVF: set up filters for vlan tags as configured */
+       for (i = 0; i < cfg->nb_pool_maps; i++) {
+               /* set vlan id in VF register and set the valid bit */
+               E1000_WRITE_REG(hw, E1000_VLVF(i), (E1000_VLVF_VLANID_ENABLE | \
+                        (cfg->pool_map[i].vlan_id & ETH_VLAN_ID_MAX) | \
+                       ((cfg->pool_map[i].pools << E1000_VLVF_POOLSEL_SHIFT ) & \
+                       E1000_VLVF_POOLSEL_MASK)));
+       }
+
+       E1000_WRITE_FLUSH(hw);
+       
+       return 0;
+}
+
+
 /*********************************************************************
  *
  *  Enable receive unit.
@@ -1605,10 +1756,20 @@ igb_dev_mq_rx_configure(struct rte_eth_dev *dev)
                /*
                * SRIOV inactive scheme
                */
-               if (dev->data->nb_rx_queues > 1)
-                       igb_rss_configure(dev);
-               else
-                       igb_rss_disable(dev);
+               switch (dev->data->dev_conf.rxmode.mq_mode) {
+                       case ETH_MQ_RX_RSS:
+                               igb_rss_configure(dev);
+                               break;
+                       case ETH_MQ_RX_VMDQ_ONLY:
+                               /*Configure general VMDQ only RX parameters*/
+                               igb_vmdq_rx_hw_configure(dev); 
+                               break;
+                       case ETH_MQ_RX_NONE:
+                               /* if mq_mode is none, disable rss mode.*/
+                       default: 
+                               igb_rss_disable(dev);
+                               break;
+               }
        }
  
        return 0;
@@ -1689,8 +1850,7 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                /*
                 * Configure RX buffer size.
                 */
-               mbp_priv = (struct rte_pktmbuf_pool_private *)
-                       ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
+               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
                buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
                                       RTE_PKTMBUF_HEADROOM);
                if (buf_size >= 1024) {
@@ -1765,6 +1925,9 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
         */
        igb_dev_mq_rx_configure(dev);
 
+       /* Update the rctl since igb_dev_mq_rx_configure may change its value */
+       rctl |= E1000_READ_REG(hw, E1000_RCTL);
+
        /*
         * Setup the Checksum Register.
         * Receive Full-Packet Checksum Offload is mutually exclusive with RSS.
@@ -1783,8 +1946,11 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
        if (dev->data->dev_conf.rxmode.hw_strip_crc) {
                rctl |= E1000_RCTL_SECRC; /* Strip Ethernet CRC. */
 
-               /* set STRCRC bit in all queues for Powerville/Springville */
-               if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
+               /* set STRCRC bit in all queues */
+               if (hw->mac.type == e1000_i350 ||
+                   hw->mac.type == e1000_i210 ||
+                   hw->mac.type == e1000_i211 ||
+                   hw->mac.type == e1000_i354) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                                rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
@@ -1796,8 +1962,11 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
        } else {
                rctl &= ~E1000_RCTL_SECRC; /* Do not Strip Ethernet CRC. */
 
-               /* clear STRCRC bit in all queues for Powerville/Springville */
-               if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
+               /* clear STRCRC bit in all queues */
+               if (hw->mac.type == e1000_i350 ||
+                   hw->mac.type == e1000_i210 ||
+                   hw->mac.type == e1000_i211 ||
+                   hw->mac.type == e1000_i354) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                                rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
@@ -1814,7 +1983,8 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
 
        /* Make sure VLAN Filters are off. */
-       rctl &= ~E1000_RCTL_VFE;
+       if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_VMDQ_ONLY)
+               rctl &= ~E1000_RCTL_VFE;
        /* Don't store bad packets. */
        rctl &= ~E1000_RCTL_SBP;
 
@@ -1907,6 +2077,11 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
 
        hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
+       /* setup MTU */
+       e1000_rlpml_set_vf(hw,
+               (uint16_t)(dev->data->dev_conf.rxmode.max_rx_pkt_len +
+               VLAN_TAG_SIZE));
+
        /* Configure and enable each RX queue. */
        rctl_bsize = 0;
        dev->rx_pkt_burst = eth_igb_recv_pkts;
@@ -1934,8 +2109,7 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                /*
                 * Configure RX buffer size.
                 */
-               mbp_priv = (struct rte_pktmbuf_pool_private *)
-                       ((char *)rxq->mb_pool + sizeof(struct rte_mempool));
+               mbp_priv = rte_mempool_get_priv(rxq->mb_pool);
                buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
                                       RTE_PKTMBUF_HEADROOM);
                if (buf_size >= 1024) {
@@ -1980,7 +2154,7 @@ eth_igbvf_rx_init(struct rte_eth_dev *dev)
                rxdctl &= 0xFFF00000;
                rxdctl |= (rxq->pthresh & 0x1F);
                rxdctl |= ((rxq->hthresh & 0x1F) << 8);
-               if (hw->mac.type == e1000_82576) {
+               if (hw->mac.type == e1000_vfadapt) {
                        /* 
                         * Workaround of 82576 VF Erratum
                         * force set WTHRESH to 1