net/cxgbe: implement flow query operation
[dpdk.git] / drivers / net / cxgbe / sge.c
index 54e13fb..357b485 100644 (file)
@@ -1,34 +1,6 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2014-2015 Chelsio Communications.
- *   All rights reserved.
- *
- *   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
- *       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
- *       distribution.
- *     * Neither the name of Chelsio Communications 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
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2014-2018 Chelsio Communications.
+ * All rights reserved.
  */
 
 #include <sys/queue.h>
@@ -82,6 +54,11 @@ static inline void ship_tx_pkt_coalesce_wr(struct adapter *adap,
  */
 #define MAX_IMM_TX_PKT_LEN 256
 
+/*
+ * Max size of a WR sent through a control Tx queue.
+ */
+#define MAX_CTRL_WR_LEN SGE_MAX_WR_LEN
+
 /*
  * Rx buffer sizes for "usembufs" Free List buffers (one ingress packet
  * per mbuf buffer).  We currently only support two sizes for 1500- and
@@ -389,7 +366,8 @@ static unsigned int refill_fl_usembufs(struct adapter *adap, struct sge_fl *q,
        struct rte_mbuf *buf_bulk[n];
        int ret, i;
        struct rte_pktmbuf_pool_private *mbp_priv;
-       u8 jumbo_en = rxq->rspq.eth_dev->data->dev_conf.rxmode.jumbo_frame;
+       u8 jumbo_en = rxq->rspq.eth_dev->data->dev_conf.rxmode.offloads &
+               DEV_RX_OFFLOAD_JUMBO_FRAME;
 
        /* Use jumbo mtu buffers if mbuf data room size can fit jumbo data. */
        mbp_priv = rte_mempool_get_priv(rxq->rspq.mb_pool);
@@ -1326,6 +1304,126 @@ out_free:
        return 0;
 }
 
+/**
+ * reclaim_completed_tx_imm - reclaim completed control-queue Tx descs
+ * @q: the SGE control Tx queue
+ *
+ * This is a variant of reclaim_completed_tx() that is used for Tx queues
+ * that send only immediate data (presently just the control queues) and
+ * thus do not have any mbufs to release.
+ */
+static inline void reclaim_completed_tx_imm(struct sge_txq *q)
+{
+       int hw_cidx = ntohs(q->stat->cidx);
+       int reclaim = hw_cidx - q->cidx;
+
+       if (reclaim < 0)
+               reclaim += q->size;
+
+       q->in_use -= reclaim;
+       q->cidx = hw_cidx;
+}
+
+/**
+ * is_imm - check whether a packet can be sent as immediate data
+ * @mbuf: the packet
+ *
+ * Returns true if a packet can be sent as a WR with immediate data.
+ */
+static inline int is_imm(const struct rte_mbuf *mbuf)
+{
+       return mbuf->pkt_len <= MAX_CTRL_WR_LEN;
+}
+
+/**
+ * inline_tx_mbuf: inline a packet's data into TX descriptors
+ * @q: the TX queue where the packet will be inlined
+ * @from: pointer to data portion of packet
+ * @to: pointer after cpl where data has to be inlined
+ * @len: length of data to inline
+ *
+ * Inline a packet's contents directly to TX descriptors, starting at
+ * the given position within the TX DMA ring.
+ * Most of the complexity of this operation is dealing with wrap arounds
+ * in the middle of the packet we want to inline.
+ */
+static void inline_tx_mbuf(const struct sge_txq *q, caddr_t from, caddr_t *to,
+                          int len)
+{
+       int left = RTE_PTR_DIFF(q->stat, *to);
+
+       if (likely((uintptr_t)*to + len <= (uintptr_t)q->stat)) {
+               rte_memcpy(*to, from, len);
+               *to = RTE_PTR_ADD(*to, len);
+       } else {
+               rte_memcpy(*to, from, left);
+               from = RTE_PTR_ADD(from, left);
+               left = len - left;
+               rte_memcpy((void *)q->desc, from, left);
+               *to = RTE_PTR_ADD((void *)q->desc, left);
+       }
+}
+
+/**
+ * ctrl_xmit - send a packet through an SGE control Tx queue
+ * @q: the control queue
+ * @mbuf: the packet
+ *
+ * Send a packet through an SGE control Tx queue.  Packets sent through
+ * a control queue must fit entirely as immediate data.
+ */
+static int ctrl_xmit(struct sge_ctrl_txq *q, struct rte_mbuf *mbuf)
+{
+       unsigned int ndesc;
+       struct fw_wr_hdr *wr;
+       caddr_t dst;
+
+       if (unlikely(!is_imm(mbuf))) {
+               WARN_ON(1);
+               rte_pktmbuf_free(mbuf);
+               return -1;
+       }
+
+       reclaim_completed_tx_imm(&q->q);
+       ndesc = DIV_ROUND_UP(mbuf->pkt_len, sizeof(struct tx_desc));
+       t4_os_lock(&q->ctrlq_lock);
+
+       q->full = txq_avail(&q->q) < ndesc ? 1 : 0;
+       if (unlikely(q->full)) {
+               t4_os_unlock(&q->ctrlq_lock);
+               return -1;
+       }
+
+       wr = (struct fw_wr_hdr *)&q->q.desc[q->q.pidx];
+       dst = (void *)wr;
+       inline_tx_mbuf(&q->q, rte_pktmbuf_mtod(mbuf, caddr_t),
+                      &dst, mbuf->data_len);
+
+       txq_advance(&q->q, ndesc);
+       if (unlikely(txq_avail(&q->q) < 64))
+               wr->lo |= htonl(F_FW_WR_EQUEQ);
+
+       q->txp++;
+
+       ring_tx_db(q->adapter, &q->q);
+       t4_os_unlock(&q->ctrlq_lock);
+
+       rte_pktmbuf_free(mbuf);
+       return 0;
+}
+
+/**
+ * t4_mgmt_tx - send a management message
+ * @q: the control queue
+ * @mbuf: the packet containing the management message
+ *
+ * Send a management message through control queue.
+ */
+int t4_mgmt_tx(struct sge_ctrl_txq *q, struct rte_mbuf *mbuf)
+{
+       return ctrl_xmit(q, mbuf);
+}
+
 /**
  * alloc_ring - allocate resources for an SGE descriptor ring
  * @dev: the PCI device's core device
@@ -1372,7 +1470,8 @@ static void *alloc_ring(size_t nelem, size_t elem_size,
         * handle the maximum ring size is allocated in order to allow for
         * resizing in later calls to the queue setup function.
         */
-       tz = rte_memzone_reserve_aligned(z_name, len, socket_id, 0, 4096);
+       tz = rte_memzone_reserve_aligned(z_name, len, socket_id,
+                       RTE_MEMZONE_IOVA_CONTIG, 4096);
        if (!tz)
                return NULL;
 
@@ -1624,7 +1723,8 @@ static int process_responses(struct sge_rspq *q, int budget,
                                }
 
                                if (cpl->vlan_ex) {
-                                       pkt->ol_flags |= PKT_RX_VLAN;
+                                       pkt->ol_flags |= PKT_RX_VLAN |
+                                                        PKT_RX_VLAN_STRIPPED;
                                        pkt->vlan_tci = ntohs(cpl->vlan);
                                }
 
@@ -2105,6 +2205,64 @@ int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq,
        return 0;
 }
 
+int t4_sge_alloc_ctrl_txq(struct adapter *adap, struct sge_ctrl_txq *txq,
+                         struct rte_eth_dev *eth_dev, uint16_t queue_id,
+                         unsigned int iqid, int socket_id)
+{
+       int ret, nentries;
+       struct fw_eq_ctrl_cmd c;
+       struct sge *s = &adap->sge;
+       struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
+       char z_name[RTE_MEMZONE_NAMESIZE];
+       char z_name_sw[RTE_MEMZONE_NAMESIZE];
+
+       /* Add status entries */
+       nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);
+
+       snprintf(z_name, sizeof(z_name), "%s_%s_%d_%d",
+                eth_dev->device->driver->name, "ctrl_tx_ring",
+                eth_dev->data->port_id, queue_id);
+       snprintf(z_name_sw, sizeof(z_name_sw), "%s_sw_ring", z_name);
+
+       txq->q.desc = alloc_ring(txq->q.size, sizeof(struct tx_desc),
+                                0, &txq->q.phys_addr,
+                                NULL, 0, queue_id,
+                                socket_id, z_name, z_name_sw);
+       if (!txq->q.desc)
+               return -ENOMEM;
+
+       memset(&c, 0, sizeof(c));
+       c.op_to_vfn = htonl(V_FW_CMD_OP(FW_EQ_CTRL_CMD) | F_FW_CMD_REQUEST |
+                           F_FW_CMD_WRITE | F_FW_CMD_EXEC |
+                           V_FW_EQ_CTRL_CMD_PFN(adap->pf) |
+                           V_FW_EQ_CTRL_CMD_VFN(0));
+       c.alloc_to_len16 = htonl(F_FW_EQ_CTRL_CMD_ALLOC |
+                                F_FW_EQ_CTRL_CMD_EQSTART | (sizeof(c) / 16));
+       c.cmpliqid_eqid = htonl(V_FW_EQ_CTRL_CMD_CMPLIQID(0));
+       c.physeqid_pkd = htonl(0);
+       c.fetchszm_to_iqid =
+               htonl(V_FW_EQ_CTRL_CMD_HOSTFCMODE(X_HOSTFCMODE_NONE) |
+                     V_FW_EQ_CTRL_CMD_PCIECHN(pi->tx_chan) |
+                     F_FW_EQ_CTRL_CMD_FETCHRO | V_FW_EQ_CTRL_CMD_IQID(iqid));
+       c.dcaen_to_eqsize =
+               htonl(V_FW_EQ_CTRL_CMD_FBMIN(X_FETCHBURSTMIN_64B) |
+                     V_FW_EQ_CTRL_CMD_FBMAX(X_FETCHBURSTMAX_512B) |
+                     V_FW_EQ_CTRL_CMD_EQSIZE(nentries));
+       c.eqaddr = cpu_to_be64(txq->q.phys_addr);
+
+       ret = t4_wr_mbox(adap, adap->mbox, &c, sizeof(c), &c);
+       if (ret) {
+               txq->q.desc = NULL;
+               return ret;
+       }
+
+       init_txq(adap, &txq->q, G_FW_EQ_CTRL_CMD_EQID(ntohl(c.cmpliqid_eqid)),
+                G_FW_EQ_CTRL_CMD_EQID(ntohl(c. physeqid_pkd)));
+       txq->adapter = adap;
+       txq->full = 0;
+       return 0;
+}
+
 static void free_txq(struct sge_txq *q)
 {
        q->cntxt_id = 0;
@@ -2199,7 +2357,7 @@ void t4_sge_tx_monitor_stop(struct adapter *adap)
  */
 void t4_free_sge_resources(struct adapter *adap)
 {
-       int i;
+       unsigned int i;
        struct sge_eth_rxq *rxq = &adap->sge.ethrxq[0];
        struct sge_eth_txq *txq = &adap->sge.ethtxq[0];
 
@@ -2216,6 +2374,18 @@ void t4_free_sge_resources(struct adapter *adap)
                }
        }
 
+       /* clean up control Tx queues */
+       for (i = 0; i < ARRAY_SIZE(adap->sge.ctrlq); i++) {
+               struct sge_ctrl_txq *cq = &adap->sge.ctrlq[i];
+
+               if (cq->q.desc) {
+                       reclaim_completed_tx_imm(&cq->q);
+                       t4_ctrl_eq_free(adap, adap->mbox, adap->pf, 0,
+                                       cq->q.cntxt_id);
+                       free_txq(&cq->q);
+               }
+       }
+
        if (adap->sge.fw_evtq.desc)
                free_rspq_fl(adap, &adap->sge.fw_evtq, NULL);
 }