mbuf: add rte prefix to offload flags
[dpdk.git] / drivers / net / e1000 / em_rxtx.c
index ed78880..a5dfe16 100644 (file)
@@ -18,6 +18,7 @@
 #include <rte_log.h>
 #include <rte_debug.h>
 #include <rte_pci.h>
+#include <rte_bus_pci.h>
 #include <rte_memory.h>
 #include <rte_memcpy.h>
 #include <rte_memzone.h>
@@ -31,7 +32,7 @@
 #include <rte_malloc.h>
 #include <rte_mbuf.h>
 #include <rte_ether.h>
-#include <rte_ethdev_driver.h>
+#include <ethdev_driver.h>
 #include <rte_prefetch.h>
 #include <rte_ip.h>
 #include <rte_udp.h>
 
 #define E1000_RXDCTL_GRAN      0x01000000 /* RXDCTL Granularity */
 
-#define E1000_TX_OFFLOAD_MASK ( \
-               PKT_TX_IPV6 |           \
-               PKT_TX_IPV4 |           \
-               PKT_TX_IP_CKSUM |       \
-               PKT_TX_L4_MASK |        \
-               PKT_TX_VLAN_PKT)
+#define E1000_TX_OFFLOAD_MASK (RTE_MBUF_F_TX_IPV6 |           \
+               RTE_MBUF_F_TX_IPV4 |           \
+               RTE_MBUF_F_TX_IP_CKSUM |       \
+               RTE_MBUF_F_TX_L4_MASK |        \
+               RTE_MBUF_F_TX_VLAN_PKT)
 
 #define E1000_TX_OFFLOAD_NOTSUP_MASK \
-               (PKT_TX_OFFLOAD_MASK ^ E1000_TX_OFFLOAD_MASK)
+               (RTE_MBUF_F_TX_OFFLOAD_MASK ^ E1000_TX_OFFLOAD_MASK)
+
+/* PCI offset for querying configuration status register */
+#define PCI_CFG_STATUS_REG                 0x06
+#define FLUSH_DESC_REQUIRED               0x100
+
 
 /**
  * Structure associated with each descriptor of the RX ring of a RX queue.
@@ -228,7 +233,7 @@ em_set_xmit_ctx(struct em_tx_queue* txq,
         * When doing checksum or TCP segmentation with IPv6 headers,
         * IPCSE field should be set t0 0.
         */
-       if (flags & PKT_TX_IP_CKSUM) {
+       if (flags & RTE_MBUF_F_TX_IP_CKSUM) {
                ctx.lower_setup.ip_fields.ipcse =
                        (uint16_t)rte_cpu_to_le_16(ipcse - 1);
                cmd_len |= E1000_TXD_CMD_IP;
@@ -241,13 +246,13 @@ em_set_xmit_ctx(struct em_tx_queue* txq,
        ctx.upper_setup.tcp_fields.tucss = (uint8_t)ipcse;
        ctx.upper_setup.tcp_fields.tucse = 0;
 
-       switch (flags & PKT_TX_L4_MASK) {
-       case PKT_TX_UDP_CKSUM:
+       switch (flags & RTE_MBUF_F_TX_L4_MASK) {
+       case RTE_MBUF_F_TX_UDP_CKSUM:
                ctx.upper_setup.tcp_fields.tucso = (uint8_t)(ipcse +
-                               offsetof(struct udp_hdr, dgram_cksum));
+                               offsetof(struct rte_udp_hdr, dgram_cksum));
                cmp_mask |= TX_MACIP_LEN_CMP_MASK;
                break;
-       case PKT_TX_TCP_CKSUM:
+       case RTE_MBUF_F_TX_TCP_CKSUM:
                ctx.upper_setup.tcp_fields.tucso = (uint8_t)(ipcse +
                                offsetof(struct rte_tcp_hdr, cksum));
                cmd_len |= E1000_TXD_CMD_TCP;
@@ -305,10 +310,10 @@ em_xmit_cleanup(struct em_tx_queue *txq)
        desc_to_clean_to = sw_ring[desc_to_clean_to].last_id;
        if (! (txr[desc_to_clean_to].upper.fields.status & E1000_TXD_STAT_DD))
        {
-               PMD_TX_FREE_LOG(DEBUG,
-                               "TX descriptor %4u is not done"
-                               "(port=%d queue=%d)", desc_to_clean_to,
-                               txq->port_id, txq->queue_id);
+               PMD_TX_LOG(DEBUG,
+                          "TX descriptor %4u is not done"
+                          "(port=%d queue=%d)", desc_to_clean_to,
+                          txq->port_id, txq->queue_id);
                /* Failed to clean any descriptors, better luck next time */
                return -(1);
        }
@@ -321,11 +326,11 @@ em_xmit_cleanup(struct em_tx_queue *txq)
                nb_tx_to_clean = (uint16_t)(desc_to_clean_to -
                                                last_desc_cleaned);
 
-       PMD_TX_FREE_LOG(DEBUG,
-                       "Cleaning %4u TX descriptors: %4u to %4u "
-                       "(port=%d queue=%d)", nb_tx_to_clean,
-                       last_desc_cleaned, desc_to_clean_to, txq->port_id,
-                       txq->queue_id);
+       PMD_TX_LOG(DEBUG,
+                  "Cleaning %4u TX descriptors: %4u to %4u "
+                  "(port=%d queue=%d)", nb_tx_to_clean,
+                  last_desc_cleaned, desc_to_clean_to, txq->port_id,
+                  txq->queue_id);
 
        /*
         * The last descriptor to clean is done, so that means all the
@@ -350,8 +355,8 @@ tx_desc_cksum_flags_to_upper(uint64_t ol_flags)
        static const uint32_t l3_olinfo[2] = {0, E1000_TXD_POPTS_IXSM << 8};
        uint32_t tmp;
 
-       tmp = l4_olinfo[(ol_flags & PKT_TX_L4_MASK) != PKT_TX_L4_NO_CKSUM];
-       tmp |= l3_olinfo[(ol_flags & PKT_TX_IP_CKSUM) != 0];
+       tmp = l4_olinfo[(ol_flags & RTE_MBUF_F_TX_L4_MASK) != RTE_MBUF_F_TX_L4_NO_CKSUM];
+       tmp |= l3_olinfo[(ol_flags & RTE_MBUF_F_TX_IP_CKSUM) != 0];
        return tmp;
 }
 
@@ -404,7 +409,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                ol_flags = tx_pkt->ol_flags;
 
                /* If hardware offload required */
-               tx_ol_req = (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK));
+               tx_ol_req = (ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_L4_MASK));
                if (tx_ol_req) {
                        hdrlen.f.vlan_tci = tx_pkt->vlan_tci;
                        hdrlen.f.l2_len = tx_pkt->l2_len;
@@ -451,11 +456,11 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                 * nb_used better be less than or equal to txq->tx_rs_thresh
                 */
                while (unlikely (nb_used > txq->nb_tx_free)) {
-                       PMD_TX_FREE_LOG(DEBUG, "Not enough free TX descriptors "
-                                       "nb_used=%4u nb_free=%4u "
-                                       "(port=%d queue=%d)",
-                                       nb_used, txq->nb_tx_free,
-                                       txq->port_id, txq->queue_id);
+                       PMD_TX_LOG(DEBUG, "Not enough free TX descriptors "
+                                  "nb_used=%4u nb_free=%4u "
+                                  "(port=%d queue=%d)",
+                                  nb_used, txq->nb_tx_free,
+                                  txq->port_id, txq->queue_id);
 
                        if (em_xmit_cleanup(txq) != 0) {
                                /* Could not clean any descriptors */
@@ -500,7 +505,7 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
                popts_spec = 0;
 
                /* Set VLAN Tag offload fields. */
-               if (ol_flags & PKT_TX_VLAN_PKT) {
+               if (ol_flags & RTE_MBUF_F_TX_VLAN_PKT) {
                        cmd_type_len |= E1000_TXD_CMD_VLE;
                        popts_spec = tx_pkt->vlan_tci << E1000_TXD_VLAN_SHIFT;
                }
@@ -573,10 +578,10 @@ eth_em_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 
                /* Set RS bit only on threshold packets' last descriptor */
                if (txq->nb_tx_used >= txq->tx_rs_thresh) {
-                       PMD_TX_FREE_LOG(DEBUG,
-                                       "Setting RS bit on TXD id=%4u "
-                                       "(port=%d queue=%d)",
-                                       tx_last, txq->port_id, txq->queue_id);
+                       PMD_TX_LOG(DEBUG,
+                                  "Setting RS bit on TXD id=%4u "
+                                  "(port=%d queue=%d)",
+                                  tx_last, txq->port_id, txq->queue_id);
 
                        cmd_type_len |= E1000_TXD_CMD_RS;
 
@@ -616,20 +621,20 @@ eth_em_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
                m = tx_pkts[i];
 
                if (m->ol_flags & E1000_TX_OFFLOAD_NOTSUP_MASK) {
-                       rte_errno = -ENOTSUP;
+                       rte_errno = ENOTSUP;
                        return i;
                }
 
-#ifdef RTE_LIBRTE_ETHDEV_DEBUG
+#ifdef RTE_ETHDEV_DEBUG_TX
                ret = rte_validate_tx_offload(m);
                if (ret != 0) {
-                       rte_errno = ret;
+                       rte_errno = -ret;
                        return i;
                }
 #endif
                ret = rte_net_intel_cksum_prepare(m);
                if (ret != 0) {
-                       rte_errno = ret;
+                       rte_errno = -ret;
                        return i;
                }
        }
@@ -650,7 +655,7 @@ rx_desc_status_to_pkt_flags(uint32_t rx_status)
 
        /* Check if VLAN present */
        pkt_flags = ((rx_status & E1000_RXD_STAT_VP) ?
-               PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED : 0);
+               RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_RX_VLAN_STRIPPED : 0);
 
        return pkt_flags;
 }
@@ -661,9 +666,9 @@ rx_desc_error_to_pkt_flags(uint32_t rx_error)
        uint64_t pkt_flags = 0;
 
        if (rx_error & E1000_RXD_ERR_IPE)
-               pkt_flags |= PKT_RX_IP_CKSUM_BAD;
+               pkt_flags |= RTE_MBUF_F_RX_IP_CKSUM_BAD;
        if (rx_error & E1000_RXD_ERR_TCPE)
-               pkt_flags |= PKT_RX_L4_CKSUM_BAD;
+               pkt_flags |= RTE_MBUF_F_RX_L4_CKSUM_BAD;
        return pkt_flags;
 }
 
@@ -805,7 +810,7 @@ eth_em_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                rxm->ol_flags = rxm->ol_flags |
                                rx_desc_error_to_pkt_flags(rxd.errors);
 
-               /* Only valid if PKT_RX_VLAN set in pkt_flags */
+               /* Only valid if RTE_MBUF_F_RX_VLAN set in pkt_flags */
                rxm->vlan_tci = rte_le_to_cpu_16(rxd.special);
 
                /*
@@ -1031,7 +1036,7 @@ eth_em_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
                first_seg->ol_flags = first_seg->ol_flags |
                                        rx_desc_error_to_pkt_flags(rxd.errors);
 
-               /* Only valid if PKT_RX_VLAN set in pkt_flags */
+               /* Only valid if RTE_MBUF_F_RX_VLAN set in pkt_flags */
                rxm->vlan_tci = rte_le_to_cpu_16(rxd.special);
 
                /* Prefetch data of first segment, if configured to do so. */
@@ -1605,12 +1610,14 @@ em_dev_free_queues(struct rte_eth_dev *dev)
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                eth_em_rx_queue_release(dev->data->rx_queues[i]);
                dev->data->rx_queues[i] = NULL;
+               rte_eth_dma_zone_free(dev, "rx_ring", i);
        }
        dev->data->nb_rx_queues = 0;
 
        for (i = 0; i < dev->data->nb_tx_queues; i++) {
                eth_em_tx_queue_release(dev->data->tx_queues[i]);
                dev->data->tx_queues[i] = NULL;
+               rte_eth_dma_zone_free(dev, "tx_ring", i);
        }
        dev->data->nb_tx_queues = 0;
 }
@@ -1964,6 +1971,22 @@ eth_em_tx_init(struct rte_eth_dev *dev)
        tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
 
+       /* SPT and CNP Si errata workaround to avoid data corruption */
+       if (hw->mac.type == e1000_pch_spt) {
+               uint32_t reg_val;
+               reg_val = E1000_READ_REG(hw, E1000_IOSFPC);
+               reg_val |= E1000_RCTL_RDMTS_HEX;
+               E1000_WRITE_REG(hw, E1000_IOSFPC, reg_val);
+
+               /* Dropping the number of outstanding requests from
+                * 3 to 2 in order to avoid a buffer overrun.
+                */
+               reg_val = E1000_READ_REG(hw, E1000_TARC(0));
+               reg_val &= ~E1000_TARC0_CB_MULTIQ_3_REQ;
+               reg_val |= E1000_TARC0_CB_MULTIQ_2_REQ;
+               E1000_WRITE_REG(hw, E1000_TARC(0), reg_val);
+       }
+
        /* This write will effectively turn on the transmit unit. */
        E1000_WRITE_REG(hw, E1000_TCTL, tctl);
 }
@@ -2000,3 +2023,119 @@ em_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
        qinfo->conf.tx_rs_thresh = txq->tx_rs_thresh;
        qinfo->conf.offloads = txq->offloads;
 }
+
+static void
+e1000_flush_tx_ring(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       volatile struct e1000_data_desc *tx_desc;
+       volatile uint32_t *tdt_reg_addr;
+       uint32_t tdt, tctl, txd_lower = E1000_TXD_CMD_IFCS;
+       uint16_t size = 512;
+       struct em_tx_queue *txq;
+       int i;
+
+       if (dev->data->tx_queues == NULL)
+               return;
+       tctl = E1000_READ_REG(hw, E1000_TCTL);
+       E1000_WRITE_REG(hw, E1000_TCTL, tctl | E1000_TCTL_EN);
+       for (i = 0; i < dev->data->nb_tx_queues &&
+               i < E1000_I219_MAX_TX_QUEUE_NUM; i++) {
+               txq = dev->data->tx_queues[i];
+               tdt = E1000_READ_REG(hw, E1000_TDT(i));
+               if (tdt != txq->tx_tail)
+                       return;
+               tx_desc = &txq->tx_ring[txq->tx_tail];
+               tx_desc->buffer_addr = rte_cpu_to_le_64(txq->tx_ring_phys_addr);
+               tx_desc->lower.data = rte_cpu_to_le_32(txd_lower | size);
+               tx_desc->upper.data = 0;
+
+               rte_io_wmb();
+               txq->tx_tail++;
+               if (txq->tx_tail == txq->nb_tx_desc)
+                       txq->tx_tail = 0;
+               tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(i));
+               E1000_PCI_REG_WRITE(tdt_reg_addr, txq->tx_tail);
+               usec_delay(250);
+       }
+}
+
+static void
+e1000_flush_rx_ring(struct rte_eth_dev *dev)
+{
+       uint32_t rctl, rxdctl;
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       int i;
+
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
+       E1000_WRITE_FLUSH(hw);
+       usec_delay(150);
+
+       for (i = 0; i < dev->data->nb_rx_queues &&
+               i < E1000_I219_MAX_RX_QUEUE_NUM; i++) {
+               rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
+               /* zero the lower 14 bits (prefetch and host thresholds) */
+               rxdctl &= 0xffffc000;
+
+               /* update thresholds: prefetch threshold to 31,
+                * host threshold to 1 and make sure the granularity
+                * is "descriptors" and not "cache lines"
+                */
+               rxdctl |= (0x1F | (1UL << 8) | E1000_RXDCTL_THRESH_UNIT_DESC);
+
+               E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
+       }
+       /* momentarily enable the RX ring for the changes to take effect */
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl | E1000_RCTL_EN);
+       E1000_WRITE_FLUSH(hw);
+       usec_delay(150);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
+}
+
+/**
+ * em_flush_desc_rings - remove all descriptors from the descriptor rings
+ *
+ * In i219, the descriptor rings must be emptied before resetting/closing the
+ * HW. Failure to do this will cause the HW to enter a unit hang state which
+ * can only be released by PCI reset on the device
+ *
+ */
+
+void
+em_flush_desc_rings(struct rte_eth_dev *dev)
+{
+       uint32_t fextnvm11, tdlen;
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       uint16_t pci_cfg_status = 0;
+       int ret;
+
+       fextnvm11 = E1000_READ_REG(hw, E1000_FEXTNVM11);
+       E1000_WRITE_REG(hw, E1000_FEXTNVM11,
+                       fextnvm11 | E1000_FEXTNVM11_DISABLE_MULR_FIX);
+       tdlen = E1000_READ_REG(hw, E1000_TDLEN(0));
+       ret = rte_pci_read_config(pci_dev, &pci_cfg_status,
+                  sizeof(pci_cfg_status), PCI_CFG_STATUS_REG);
+       if (ret < 0) {
+               PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x",
+                           PCI_CFG_STATUS_REG);
+               return;
+       }
+
+       /* do nothing if we're not in faulty state, or if the queue is empty */
+       if ((pci_cfg_status & FLUSH_DESC_REQUIRED) && tdlen) {
+               /* flush desc ring */
+               e1000_flush_tx_ring(dev);
+               ret = rte_pci_read_config(pci_dev, &pci_cfg_status,
+                               sizeof(pci_cfg_status), PCI_CFG_STATUS_REG);
+               if (ret < 0) {
+                       PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x",
+                                       PCI_CFG_STATUS_REG);
+                       return;
+               }
+
+               if (pci_cfg_status & FLUSH_DESC_REQUIRED)
+                       e1000_flush_rx_ring(dev);
+       }
+}