cxgb4: parse and configure TC-MQPRIO offload
authorRahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Thu, 7 Nov 2019 15:59:06 +0000 (21:29 +0530)
committerDavid S. Miller <davem@davemloft.net>
Thu, 7 Nov 2019 18:41:59 +0000 (10:41 -0800)
Add logic for validation and configuration of TC-MQPRIO Qdisc
offload. Also, add support to manage EOSW_TXQ, which have 1-to-1
mapping with EOTIDs, and expose them to network stack.

Move common skb validation in Tx path to a separate function and
add minimal Tx path for ETHOFLD. Update Tx queue selection to return
normal NIC Txq to send traffic pattern that can't go through ETHOFLD
Tx path.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/chelsio/cxgb4/Makefile
drivers/net/ethernet/chelsio/cxgb4/cxgb4.h
drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.c [new file with mode: 0644]
drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.h [new file with mode: 0644]
drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h
drivers/net/ethernet/chelsio/cxgb4/sge.c

index 20390f6afbb4da8fd4f008f4a7fc5a923ef84404..49a19308073bb0fec9da859d15fd6067b97c7b99 100644 (file)
@@ -8,7 +8,7 @@ obj-$(CONFIG_CHELSIO_T4) += cxgb4.o
 cxgb4-objs := cxgb4_main.o l2t.o smt.o t4_hw.o sge.o clip_tbl.o cxgb4_ethtool.o \
              cxgb4_uld.o srq.o sched.o cxgb4_filter.o cxgb4_tc_u32.o \
              cxgb4_ptp.o cxgb4_tc_flower.o cxgb4_cudbg.o cxgb4_mps.o \
-             cudbg_common.o cudbg_lib.o cudbg_zlib.o
+             cudbg_common.o cudbg_lib.o cudbg_zlib.o cxgb4_tc_mqprio.o
 cxgb4-$(CONFIG_CHELSIO_T4_DCB) +=  cxgb4_dcb.o
 cxgb4-$(CONFIG_CHELSIO_T4_FCOE) +=  cxgb4_fcoe.o
 cxgb4-$(CONFIG_DEBUG_FS) += cxgb4_debugfs.o
index dbfde3fbfce40acfae55793d9cc4639bcac2ca9c..04a81276e9c8e02b5ea0da56f11dadcf1f2058cc 100644 (file)
@@ -803,6 +803,38 @@ struct sge_uld_txq_info {
        u16 ntxq;               /* # of egress uld queues */
 };
 
+enum sge_eosw_state {
+       CXGB4_EO_STATE_CLOSED = 0, /* Not ready to accept traffic */
+};
+
+struct sge_eosw_desc {
+       struct sk_buff *skb; /* SKB to free after getting completion */
+       dma_addr_t addr[MAX_SKB_FRAGS + 1]; /* DMA mapped addresses */
+};
+
+struct sge_eosw_txq {
+       spinlock_t lock; /* Per queue lock to synchronize completions */
+       enum sge_eosw_state state; /* Current ETHOFLD State */
+       struct sge_eosw_desc *desc; /* Descriptor ring to hold packets */
+       u32 ndesc; /* Number of descriptors */
+       u32 pidx; /* Current Producer Index */
+       u32 last_pidx; /* Last successfully transmitted Producer Index */
+       u32 cidx; /* Current Consumer Index */
+       u32 last_cidx; /* Last successfully reclaimed Consumer Index */
+       u32 inuse; /* Number of packets held in ring */
+
+       u32 cred; /* Current available credits */
+       u32 ncompl; /* # of completions posted */
+       u32 last_compl; /* # of credits consumed since last completion req */
+
+       u32 eotid; /* Index into EOTID table in software */
+       u32 hwtid; /* Hardware EOTID index */
+
+       u32 hwqid; /* Underlying hardware queue index */
+       struct net_device *netdev; /* Pointer to netdevice */
+       struct tasklet_struct qresume_tsk; /* Restarts the queue */
+};
+
 struct sge {
        struct sge_eth_txq ethtxq[MAX_ETH_QSETS];
        struct sge_eth_txq ptptxq;
@@ -1044,6 +1076,9 @@ struct adapter {
 #if IS_ENABLED(CONFIG_THERMAL)
        struct ch_thermal ch_thermal;
 #endif
+
+       /* TC MQPRIO offload */
+       struct cxgb4_tc_mqprio *tc_mqprio;
 };
 
 /* Support for "sched-class" command to allow a TX Scheduling Class to be
@@ -1895,6 +1930,9 @@ int t4_i2c_rd(struct adapter *adap, unsigned int mbox, int port,
 void free_rspq_fl(struct adapter *adap, struct sge_rspq *rq, struct sge_fl *fl);
 void free_tx_desc(struct adapter *adap, struct sge_txq *q,
                  unsigned int n, bool unmap);
+void cxgb4_eosw_txq_free_desc(struct adapter *adap, struct sge_eosw_txq *txq,
+                             u32 ndesc);
+void cxgb4_ethofld_restart(unsigned long data);
 void free_txq(struct adapter *adap, struct sge_txq *q);
 void cxgb4_reclaim_completed_tx(struct adapter *adap,
                                struct sge_txq *q, bool unmap);
@@ -1955,4 +1993,6 @@ int cxgb4_update_mac_filt(struct port_info *pi, unsigned int viid,
                          bool persistent, u8 *smt_idx);
 int cxgb4_get_msix_idx_from_bmap(struct adapter *adap);
 void cxgb4_free_msix_idx_in_bmap(struct adapter *adap, u32 msix_idx);
+int cxgb_open(struct net_device *dev);
+int cxgb_close(struct net_device *dev);
 #endif /* __CXGB4_H__ */
index d8a1bd80b29372b0380a891404a21c5f2bcbadfa..fe3ea60843c33995a7dc0aca77e3b8f1043cec56 100644 (file)
@@ -65,6 +65,7 @@
 #include <linux/uaccess.h>
 #include <linux/crash_dump.h>
 #include <net/udp_tunnel.h>
+#include <net/xfrm.h>
 
 #include "cxgb4.h"
 #include "cxgb4_filter.h"
@@ -82,6 +83,7 @@
 #include "sched.h"
 #include "cxgb4_tc_u32.h"
 #include "cxgb4_tc_flower.h"
+#include "cxgb4_tc_mqprio.h"
 #include "cxgb4_ptp.h"
 #include "cxgb4_cudbg.h"
 
@@ -1117,6 +1119,18 @@ static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
        }
 #endif /* CONFIG_CHELSIO_T4_DCB */
 
+       if (dev->num_tc) {
+               struct port_info *pi = netdev2pinfo(dev);
+
+               /* Send unsupported traffic pattern to normal NIC queues. */
+               txq = netdev_pick_tx(dev, skb, sb_dev);
+               if (xfrm_offload(skb) || is_ptp_enabled(skb, dev) ||
+                   ip_hdr(skb)->protocol != IPPROTO_TCP)
+                       txq = txq % pi->nqsets;
+
+               return txq;
+       }
+
        if (select_queue) {
                txq = (skb_rx_queue_recorded(skb)
                        ? skb_get_rx_queue(skb)
@@ -2472,11 +2486,11 @@ static void cxgb_down(struct adapter *adapter)
 /*
  * net_device operations
  */
-static int cxgb_open(struct net_device *dev)
+int cxgb_open(struct net_device *dev)
 {
-       int err;
        struct port_info *pi = netdev_priv(dev);
        struct adapter *adapter = pi->adapter;
+       int err;
 
        netif_carrier_off(dev);
 
@@ -2499,7 +2513,7 @@ static int cxgb_open(struct net_device *dev)
        return err;
 }
 
-static int cxgb_close(struct net_device *dev)
+int cxgb_close(struct net_device *dev)
 {
        struct port_info *pi = netdev_priv(dev);
        struct adapter *adapter = pi->adapter;
@@ -3233,6 +3247,17 @@ static int cxgb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
        }
 }
 
+static int cxgb_setup_tc_mqprio(struct net_device *dev,
+                               struct tc_mqprio_qopt_offload *mqprio)
+{
+       struct adapter *adap = netdev2adap(dev);
+
+       if (!is_ethofld(adap) || !adap->tc_mqprio)
+               return -ENOMEM;
+
+       return cxgb4_setup_tc_mqprio(dev, mqprio);
+}
+
 static LIST_HEAD(cxgb_block_cb_list);
 
 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
@@ -3241,6 +3266,8 @@ static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
        struct port_info *pi = netdev2pinfo(dev);
 
        switch (type) {
+       case TC_SETUP_QDISC_MQPRIO:
+               return cxgb_setup_tc_mqprio(dev, type_data);
        case TC_SETUP_BLOCK:
                return flow_block_cb_setup_simple(type_data,
                                                  &cxgb_block_cb_list,
@@ -5668,6 +5695,7 @@ static void free_some_resources(struct adapter *adapter)
        kvfree(adapter->srq);
        t4_cleanup_sched(adapter);
        kvfree(adapter->tids.tid_tab);
+       cxgb4_cleanup_tc_mqprio(adapter);
        cxgb4_cleanup_tc_flower(adapter);
        cxgb4_cleanup_tc_u32(adapter);
        kfree(adapter->sge.egr_map);
@@ -6237,6 +6265,10 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
                if (cxgb4_init_tc_flower(adapter))
                        dev_warn(&pdev->dev,
                                 "could not offload tc flower, continuing\n");
+
+               if (cxgb4_init_tc_mqprio(adapter))
+                       dev_warn(&pdev->dev,
+                                "could not offload tc mqprio, continuing\n");
        }
 
        if (is_offload(adapter) || is_hashfilter(adapter)) {
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.c
new file mode 100644 (file)
index 0000000..8cdf3dc
--- /dev/null
@@ -0,0 +1,346 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (C) 2019 Chelsio Communications.  All rights reserved. */
+
+#include "cxgb4.h"
+#include "cxgb4_tc_mqprio.h"
+
+static int cxgb4_mqprio_validate(struct net_device *dev,
+                                struct tc_mqprio_qopt_offload *mqprio)
+{
+       u64 min_rate = 0, max_rate = 0, max_link_rate;
+       struct port_info *pi = netdev2pinfo(dev);
+       struct adapter *adap = netdev2adap(dev);
+       u32 qcount = 0, qoffset = 0;
+       u32 link_ok, speed, mtu;
+       int ret;
+       u8 i;
+
+       if (!mqprio->qopt.num_tc)
+               return 0;
+
+       if (mqprio->qopt.hw != TC_MQPRIO_HW_OFFLOAD_TCS) {
+               netdev_err(dev, "Only full TC hardware offload is supported\n");
+               return -EINVAL;
+       } else if (mqprio->mode != TC_MQPRIO_MODE_CHANNEL) {
+               netdev_err(dev, "Only channel mode offload is supported\n");
+               return -EINVAL;
+       } else if (mqprio->shaper != TC_MQPRIO_SHAPER_BW_RATE) {
+               netdev_err(dev, "Only bandwidth rate shaper supported\n");
+               return -EINVAL;
+       } else if (mqprio->qopt.num_tc > adap->params.nsched_cls) {
+               netdev_err(dev,
+                          "Only %u traffic classes supported by hardware\n",
+                          adap->params.nsched_cls);
+               return -ERANGE;
+       }
+
+       ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
+       if (ret) {
+               netdev_err(dev, "Failed to get link speed, ret: %d\n", ret);
+               return -EINVAL;
+       }
+
+       /* Convert from Mbps to bps */
+       max_link_rate = (u64)speed * 1000 * 1000;
+
+       for (i = 0; i < mqprio->qopt.num_tc; i++) {
+               qoffset = max_t(u16, mqprio->qopt.offset[i], qoffset);
+               qcount += mqprio->qopt.count[i];
+
+               /* Convert byte per second to bits per second */
+               min_rate += (mqprio->min_rate[i] * 8);
+               max_rate += (mqprio->max_rate[i] * 8);
+       }
+
+       if (qoffset >= adap->tids.neotids || qcount > adap->tids.neotids)
+               return -ENOMEM;
+
+       if (min_rate > max_link_rate || max_rate > max_link_rate) {
+               netdev_err(dev,
+                          "Total Min/Max (%llu/%llu) Rate > supported (%llu)\n",
+                          min_rate, max_rate, max_link_rate);
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int cxgb4_init_eosw_txq(struct net_device *dev,
+                              struct sge_eosw_txq *eosw_txq,
+                              u32 eotid, u32 hwqid)
+{
+       struct adapter *adap = netdev2adap(dev);
+       struct sge_eosw_desc *ring;
+
+       memset(eosw_txq, 0, sizeof(*eosw_txq));
+
+       ring = kcalloc(CXGB4_EOSW_TXQ_DEFAULT_DESC_NUM,
+                      sizeof(*ring), GFP_KERNEL);
+       if (!ring)
+               return -ENOMEM;
+
+       eosw_txq->desc = ring;
+       eosw_txq->ndesc = CXGB4_EOSW_TXQ_DEFAULT_DESC_NUM;
+       spin_lock_init(&eosw_txq->lock);
+       eosw_txq->state = CXGB4_EO_STATE_CLOSED;
+       eosw_txq->eotid = eotid;
+       eosw_txq->hwtid = adap->tids.eotid_base + eosw_txq->eotid;
+       eosw_txq->cred = adap->params.ofldq_wr_cred;
+       eosw_txq->hwqid = hwqid;
+       eosw_txq->netdev = dev;
+       tasklet_init(&eosw_txq->qresume_tsk, cxgb4_ethofld_restart,
+                    (unsigned long)eosw_txq);
+       return 0;
+}
+
+static void cxgb4_clean_eosw_txq(struct net_device *dev,
+                                struct sge_eosw_txq *eosw_txq)
+{
+       struct adapter *adap = netdev2adap(dev);
+
+       cxgb4_eosw_txq_free_desc(adap, eosw_txq, eosw_txq->ndesc);
+       eosw_txq->pidx = 0;
+       eosw_txq->last_pidx = 0;
+       eosw_txq->cidx = 0;
+       eosw_txq->last_cidx = 0;
+       eosw_txq->inuse = 0;
+       eosw_txq->cred = adap->params.ofldq_wr_cred;
+       eosw_txq->ncompl = 0;
+       eosw_txq->last_compl = 0;
+       eosw_txq->state = CXGB4_EO_STATE_CLOSED;
+}
+
+static void cxgb4_free_eosw_txq(struct net_device *dev,
+                               struct sge_eosw_txq *eosw_txq)
+{
+       spin_lock_bh(&eosw_txq->lock);
+       cxgb4_clean_eosw_txq(dev, eosw_txq);
+       kfree(eosw_txq->desc);
+       spin_unlock_bh(&eosw_txq->lock);
+       tasklet_kill(&eosw_txq->qresume_tsk);
+}
+
+static int cxgb4_mqprio_enable_offload(struct net_device *dev,
+                                      struct tc_mqprio_qopt_offload *mqprio)
+{
+       struct cxgb4_tc_port_mqprio *tc_port_mqprio;
+       u32 qoffset, qcount, tot_qcount, qid, hwqid;
+       struct port_info *pi = netdev2pinfo(dev);
+       struct adapter *adap = netdev2adap(dev);
+       struct sge_eosw_txq *eosw_txq;
+       int eotid, ret;
+       u16 i, j;
+
+       tc_port_mqprio = &adap->tc_mqprio->port_mqprio[pi->port_id];
+       for (i = 0; i < mqprio->qopt.num_tc; i++) {
+               qoffset = mqprio->qopt.offset[i];
+               qcount = mqprio->qopt.count[i];
+               for (j = 0; j < qcount; j++) {
+                       eotid = cxgb4_get_free_eotid(&adap->tids);
+                       if (eotid < 0) {
+                               ret = -ENOMEM;
+                               goto out_free_eotids;
+                       }
+
+                       qid = qoffset + j;
+                       hwqid = pi->first_qset + (eotid % pi->nqsets);
+                       eosw_txq = &tc_port_mqprio->eosw_txq[qid];
+                       ret = cxgb4_init_eosw_txq(dev, eosw_txq,
+                                                 eotid, hwqid);
+                       if (ret)
+                               goto out_free_eotids;
+
+                       cxgb4_alloc_eotid(&adap->tids, eotid, eosw_txq);
+               }
+       }
+
+       memcpy(&tc_port_mqprio->mqprio, mqprio,
+              sizeof(struct tc_mqprio_qopt_offload));
+
+       /* Inform the stack about the configured tc params.
+        *
+        * Set the correct queue map. If no queue count has been
+        * specified, then send the traffic through default NIC
+        * queues; instead of ETHOFLD queues.
+        */
+       ret = netdev_set_num_tc(dev, mqprio->qopt.num_tc);
+       if (ret)
+               goto out_free_eotids;
+
+       tot_qcount = pi->nqsets;
+       for (i = 0; i < mqprio->qopt.num_tc; i++) {
+               qcount = mqprio->qopt.count[i];
+               if (qcount) {
+                       qoffset = mqprio->qopt.offset[i] + pi->nqsets;
+               } else {
+                       qcount = pi->nqsets;
+                       qoffset = 0;
+               }
+
+               ret = netdev_set_tc_queue(dev, i, qcount, qoffset);
+               if (ret)
+                       goto out_reset_tc;
+
+               tot_qcount += mqprio->qopt.count[i];
+       }
+
+       ret = netif_set_real_num_tx_queues(dev, tot_qcount);
+       if (ret)
+               goto out_reset_tc;
+
+       tc_port_mqprio->state = CXGB4_MQPRIO_STATE_ACTIVE;
+       return 0;
+
+out_reset_tc:
+       netdev_reset_tc(dev);
+       i = mqprio->qopt.num_tc;
+
+out_free_eotids:
+       while (i-- > 0) {
+               qoffset = mqprio->qopt.offset[i];
+               qcount = mqprio->qopt.count[i];
+               for (j = 0; j < qcount; j++) {
+                       eosw_txq = &tc_port_mqprio->eosw_txq[qoffset + j];
+                       cxgb4_free_eotid(&adap->tids, eosw_txq->eotid);
+                       cxgb4_free_eosw_txq(dev, eosw_txq);
+               }
+       }
+
+       return ret;
+}
+
+static void cxgb4_mqprio_disable_offload(struct net_device *dev)
+{
+       struct cxgb4_tc_port_mqprio *tc_port_mqprio;
+       struct port_info *pi = netdev2pinfo(dev);
+       struct adapter *adap = netdev2adap(dev);
+       struct sge_eosw_txq *eosw_txq;
+       u32 qoffset, qcount;
+       u16 i, j;
+
+       tc_port_mqprio = &adap->tc_mqprio->port_mqprio[pi->port_id];
+       if (tc_port_mqprio->state != CXGB4_MQPRIO_STATE_ACTIVE)
+               return;
+
+       netdev_reset_tc(dev);
+       netif_set_real_num_tx_queues(dev, pi->nqsets);
+
+       for (i = 0; i < tc_port_mqprio->mqprio.qopt.num_tc; i++) {
+               qoffset = tc_port_mqprio->mqprio.qopt.offset[i];
+               qcount = tc_port_mqprio->mqprio.qopt.count[i];
+               for (j = 0; j < qcount; j++) {
+                       eosw_txq = &tc_port_mqprio->eosw_txq[qoffset + j];
+                       cxgb4_free_eotid(&adap->tids, eosw_txq->eotid);
+                       cxgb4_free_eosw_txq(dev, eosw_txq);
+               }
+       }
+
+       memset(&tc_port_mqprio->mqprio, 0,
+              sizeof(struct tc_mqprio_qopt_offload));
+
+       tc_port_mqprio->state = CXGB4_MQPRIO_STATE_DISABLED;
+}
+
+int cxgb4_setup_tc_mqprio(struct net_device *dev,
+                         struct tc_mqprio_qopt_offload *mqprio)
+{
+       bool needs_bring_up = false;
+       int ret;
+
+       ret = cxgb4_mqprio_validate(dev, mqprio);
+       if (ret)
+               return ret;
+
+       /* To configure tc params, the current allocated EOTIDs must
+        * be freed up. However, they can't be freed up if there's
+        * traffic running on the interface. So, ensure interface is
+        * down before configuring tc params.
+        */
+       if (netif_running(dev)) {
+               cxgb_close(dev);
+               needs_bring_up = true;
+       }
+
+       cxgb4_mqprio_disable_offload(dev);
+
+       /* If requested for clear, then just return since resources are
+        * already freed up by now.
+        */
+       if (!mqprio->qopt.num_tc)
+               goto out;
+
+       ret = cxgb4_mqprio_enable_offload(dev, mqprio);
+
+out:
+       if (needs_bring_up)
+               cxgb_open(dev);
+
+       return ret;
+}
+
+int cxgb4_init_tc_mqprio(struct adapter *adap)
+{
+       struct cxgb4_tc_port_mqprio *tc_port_mqprio, *port_mqprio;
+       struct cxgb4_tc_mqprio *tc_mqprio;
+       struct sge_eosw_txq *eosw_txq;
+       int ret = 0;
+       u8 i;
+
+       tc_mqprio = kzalloc(sizeof(*tc_mqprio), GFP_KERNEL);
+       if (!tc_mqprio)
+               return -ENOMEM;
+
+       tc_port_mqprio = kcalloc(adap->params.nports, sizeof(*tc_port_mqprio),
+                                GFP_KERNEL);
+       if (!tc_port_mqprio) {
+               ret = -ENOMEM;
+               goto out_free_mqprio;
+       }
+
+       tc_mqprio->port_mqprio = tc_port_mqprio;
+       for (i = 0; i < adap->params.nports; i++) {
+               port_mqprio = &tc_mqprio->port_mqprio[i];
+               eosw_txq = kcalloc(adap->tids.neotids, sizeof(*eosw_txq),
+                                  GFP_KERNEL);
+               if (!eosw_txq) {
+                       ret = -ENOMEM;
+                       goto out_free_ports;
+               }
+               port_mqprio->eosw_txq = eosw_txq;
+       }
+
+       adap->tc_mqprio = tc_mqprio;
+       return 0;
+
+out_free_ports:
+       for (i = 0; i < adap->params.nports; i++) {
+               port_mqprio = &tc_mqprio->port_mqprio[i];
+               kfree(port_mqprio->eosw_txq);
+       }
+       kfree(tc_port_mqprio);
+
+out_free_mqprio:
+       kfree(tc_mqprio);
+       return ret;
+}
+
+void cxgb4_cleanup_tc_mqprio(struct adapter *adap)
+{
+       struct cxgb4_tc_port_mqprio *port_mqprio;
+       u8 i;
+
+       if (adap->tc_mqprio) {
+               if (adap->tc_mqprio->port_mqprio) {
+                       for (i = 0; i < adap->params.nports; i++) {
+                               struct net_device *dev = adap->port[i];
+
+                               if (dev)
+                                       cxgb4_mqprio_disable_offload(dev);
+                               port_mqprio = &adap->tc_mqprio->port_mqprio[i];
+                               kfree(port_mqprio->eosw_txq);
+                       }
+                       kfree(adap->tc_mqprio->port_mqprio);
+               }
+               kfree(adap->tc_mqprio);
+       }
+}
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_tc_mqprio.h
new file mode 100644 (file)
index 0000000..125664b
--- /dev/null
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/* Copyright (C) 2019 Chelsio Communications.  All rights reserved. */
+
+#ifndef __CXGB4_TC_MQPRIO_H__
+#define __CXGB4_TC_MQPRIO_H__
+
+#include <net/pkt_cls.h>
+
+#define CXGB4_EOSW_TXQ_DEFAULT_DESC_NUM 128
+
+enum cxgb4_mqprio_state {
+       CXGB4_MQPRIO_STATE_DISABLED = 0,
+       CXGB4_MQPRIO_STATE_ACTIVE,
+};
+
+struct cxgb4_tc_port_mqprio {
+       enum cxgb4_mqprio_state state; /* Current MQPRIO offload state */
+       struct tc_mqprio_qopt_offload mqprio; /* MQPRIO offload params */
+       struct sge_eosw_txq *eosw_txq; /* Netdev SW Tx queue array */
+};
+
+struct cxgb4_tc_mqprio {
+       struct cxgb4_tc_port_mqprio *port_mqprio; /* Per port MQPRIO info */
+};
+
+int cxgb4_setup_tc_mqprio(struct net_device *dev,
+                         struct tc_mqprio_qopt_offload *mqprio);
+int cxgb4_init_tc_mqprio(struct adapter *adap);
+void cxgb4_cleanup_tc_mqprio(struct adapter *adap);
+#endif /* __CXGB4_TC_MQPRIO_H__ */
index b78feef23dc4d3b672125405f568adf88f7ceecd..861b25d28ed645890160091105d9a10ff92c8f91 100644 (file)
@@ -186,6 +186,35 @@ static inline void cxgb4_insert_tid(struct tid_info *t, void *data,
        atomic_inc(&t->conns_in_use);
 }
 
+static inline struct eotid_entry *cxgb4_lookup_eotid(struct tid_info *t,
+                                                    u32 eotid)
+{
+       return eotid < t->neotids ? &t->eotid_tab[eotid] : NULL;
+}
+
+static inline int cxgb4_get_free_eotid(struct tid_info *t)
+{
+       int eotid;
+
+       eotid = find_first_zero_bit(t->eotid_bmap, t->neotids);
+       if (eotid >= t->neotids)
+               eotid = -1;
+
+       return eotid;
+}
+
+static inline void cxgb4_alloc_eotid(struct tid_info *t, u32 eotid, void *data)
+{
+       set_bit(eotid, t->eotid_bmap);
+       t->eotid_tab[eotid].data = data;
+}
+
+static inline void cxgb4_free_eotid(struct tid_info *t, u32 eotid)
+{
+       clear_bit(eotid, t->eotid_bmap);
+       t->eotid_tab[eotid].data = NULL;
+}
+
 int cxgb4_alloc_atid(struct tid_info *t, void *data);
 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data);
 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data);
index 82d8e13fc7d2ccb414319dc1c3e1aab7cbd9a7b8..ed1418d2364f488cf0d307b4e822278e1e5526c3 100644 (file)
@@ -269,7 +269,6 @@ out_err:
 }
 EXPORT_SYMBOL(cxgb4_map_skb);
 
-#ifdef CONFIG_NEED_DMA_MAP_STATE
 static void unmap_skb(struct device *dev, const struct sk_buff *skb,
                      const dma_addr_t *addr)
 {
@@ -284,6 +283,7 @@ static void unmap_skb(struct device *dev, const struct sk_buff *skb,
                dma_unmap_page(dev, *addr++, skb_frag_size(fp), DMA_TO_DEVICE);
 }
 
+#ifdef CONFIG_NEED_DMA_MAP_STATE
 /**
  *     deferred_unmap_destructor - unmap a packet when it is freed
  *     @skb: the packet
@@ -1347,6 +1347,31 @@ int t4_sge_eth_txq_egress_update(struct adapter *adap, struct sge_eth_txq *eq,
        return reclaimed;
 }
 
+static inline int cxgb4_validate_skb(struct sk_buff *skb,
+                                    struct net_device *dev,
+                                    u32 min_pkt_len)
+{
+       u32 max_pkt_len;
+
+       /* The chip min packet length is 10 octets but some firmware
+        * commands have a minimum packet length requirement. So, play
+        * safe and reject anything shorter than @min_pkt_len.
+        */
+       if (unlikely(skb->len < min_pkt_len))
+               return -EINVAL;
+
+       /* Discard the packet if the length is greater than mtu */
+       max_pkt_len = ETH_HLEN + dev->mtu;
+
+       if (skb_vlan_tagged(skb))
+               max_pkt_len += VLAN_HLEN;
+
+       if (!skb_shinfo(skb)->gso_size && (unlikely(skb->len > max_pkt_len)))
+               return -EINVAL;
+
+       return 0;
+}
+
 /**
  *     cxgb4_eth_xmit - add a packet to an Ethernet Tx queue
  *     @skb: the packet
@@ -1356,41 +1381,24 @@ int t4_sge_eth_txq_egress_update(struct adapter *adap, struct sge_eth_txq *eq,
  */
 static netdev_tx_t cxgb4_eth_xmit(struct sk_buff *skb, struct net_device *dev)
 {
-       u32 wr_mid, ctrl0, op;
-       u64 cntrl, *end, *sgl;
-       int qidx, credits;
-       unsigned int flits, ndesc;
-       struct adapter *adap;
-       struct sge_eth_txq *q;
-       const struct port_info *pi;
+       enum cpl_tx_tnl_lso_type tnl_type = TX_TNL_TYPE_OPAQUE;
+       bool ptp_enabled = is_ptp_enabled(skb, dev);
+       dma_addr_t addr[MAX_SKB_FRAGS + 1];
+       const struct skb_shared_info *ssi;
        struct fw_eth_tx_pkt_wr *wr;
        struct cpl_tx_pkt_core *cpl;
-       const struct skb_shared_info *ssi;
-       dma_addr_t addr[MAX_SKB_FRAGS + 1];
+       int len, qidx, credits, ret;
+       const struct port_info *pi;
+       unsigned int flits, ndesc;
        bool immediate = false;
-       int len, max_pkt_len;
-       bool ptp_enabled = is_ptp_enabled(skb, dev);
+       u32 wr_mid, ctrl0, op;
+       u64 cntrl, *end, *sgl;
+       struct sge_eth_txq *q;
        unsigned int chip_ver;
-       enum cpl_tx_tnl_lso_type tnl_type = TX_TNL_TYPE_OPAQUE;
-
-#ifdef CONFIG_CHELSIO_T4_FCOE
-       int err;
-#endif /* CONFIG_CHELSIO_T4_FCOE */
-
-       /*
-        * The chip min packet length is 10 octets but play safe and reject
-        * anything shorter than an Ethernet header.
-        */
-       if (unlikely(skb->len < ETH_HLEN)) {
-out_free:      dev_kfree_skb_any(skb);
-               return NETDEV_TX_OK;
-       }
+       struct adapter *adap;
 
-       /* Discard the packet if the length is greater than mtu */
-       max_pkt_len = ETH_HLEN + dev->mtu;
-       if (skb_vlan_tagged(skb))
-               max_pkt_len += VLAN_HLEN;
-       if (!skb_shinfo(skb)->gso_size && (unlikely(skb->len > max_pkt_len)))
+       ret = cxgb4_validate_skb(skb, dev, ETH_HLEN);
+       if (ret)
                goto out_free;
 
        pi = netdev_priv(dev);
@@ -1421,8 +1429,8 @@ out_free: dev_kfree_skb_any(skb);
        cntrl = TXPKT_L4CSUM_DIS_F | TXPKT_IPCSUM_DIS_F;
 
 #ifdef CONFIG_CHELSIO_T4_FCOE
-       err = cxgb_fcoe_offload(skb, adap, pi, &cntrl);
-       if (unlikely(err == -ENOTSUPP)) {
+       ret = cxgb_fcoe_offload(skb, adap, pi, &cntrl);
+       if (unlikely(ret == -ENOTSUPP)) {
                if (ptp_enabled)
                        spin_unlock(&adap->ptp_lock);
                goto out_free;
@@ -1622,6 +1630,10 @@ out_free:        dev_kfree_skb_any(skb);
        if (ptp_enabled)
                spin_unlock(&adap->ptp_lock);
        return NETDEV_TX_OK;
+
+out_free:
+       dev_kfree_skb_any(skb);
+       return NETDEV_TX_OK;
 }
 
 /* Constants ... */
@@ -1710,32 +1722,25 @@ static netdev_tx_t cxgb4_vf_eth_xmit(struct sk_buff *skb,
        dma_addr_t addr[MAX_SKB_FRAGS + 1];
        const struct skb_shared_info *ssi;
        struct fw_eth_tx_pkt_vm_wr *wr;
-       int qidx, credits, max_pkt_len;
        struct cpl_tx_pkt_core *cpl;
        const struct port_info *pi;
        unsigned int flits, ndesc;
        struct sge_eth_txq *txq;
        struct adapter *adapter;
+       int qidx, credits, ret;
+       size_t fw_hdr_copy_len;
        u64 cntrl, *end;
        u32 wr_mid;
-       const size_t fw_hdr_copy_len = sizeof(wr->ethmacdst) +
-                                      sizeof(wr->ethmacsrc) +
-                                      sizeof(wr->ethtype) +
-                                      sizeof(wr->vlantci);
 
        /* The chip minimum packet length is 10 octets but the firmware
         * command that we are using requires that we copy the Ethernet header
         * (including the VLAN tag) into the header so we reject anything
         * smaller than that ...
         */
-       if (unlikely(skb->len < fw_hdr_copy_len))
-               goto out_free;
-
-       /* Discard the packet if the length is greater than mtu */
-       max_pkt_len = ETH_HLEN + dev->mtu;
-       if (skb_vlan_tag_present(skb))
-               max_pkt_len += VLAN_HLEN;
-       if (!skb_shinfo(skb)->gso_size && (unlikely(skb->len > max_pkt_len)))
+       fw_hdr_copy_len = sizeof(wr->ethmacdst) + sizeof(wr->ethmacsrc) +
+                         sizeof(wr->ethtype) + sizeof(wr->vlantci);
+       ret = cxgb4_validate_skb(skb, dev, fw_hdr_copy_len);
+       if (ret)
                goto out_free;
 
        /* Figure out which TX Queue we're going to use. */
@@ -1991,13 +1996,62 @@ out_free:
        return NETDEV_TX_OK;
 }
 
+static inline void eosw_txq_advance_index(u32 *idx, u32 n, u32 max)
+{
+       u32 val = *idx + n;
+
+       if (val >= max)
+               val -= max;
+
+       *idx = val;
+}
+
+void cxgb4_eosw_txq_free_desc(struct adapter *adap,
+                             struct sge_eosw_txq *eosw_txq, u32 ndesc)
+{
+       struct sge_eosw_desc *d;
+
+       d = &eosw_txq->desc[eosw_txq->last_cidx];
+       while (ndesc--) {
+               if (d->skb) {
+                       if (d->addr[0]) {
+                               unmap_skb(adap->pdev_dev, d->skb, d->addr);
+                               memset(d->addr, 0, sizeof(d->addr));
+                       }
+                       dev_consume_skb_any(d->skb);
+                       d->skb = NULL;
+               }
+               eosw_txq_advance_index(&eosw_txq->last_cidx, 1,
+                                      eosw_txq->ndesc);
+               d = &eosw_txq->desc[eosw_txq->last_cidx];
+       }
+}
+
+static netdev_tx_t cxgb4_ethofld_xmit(struct sk_buff *skb,
+                                     struct net_device *dev)
+{
+       int ret;
+
+       ret = cxgb4_validate_skb(skb, dev, ETH_HLEN);
+       if (ret)
+               goto out_free;
+
+out_free:
+       dev_kfree_skb_any(skb);
+       return NETDEV_TX_OK;
+}
+
 netdev_tx_t t4_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct port_info *pi = netdev_priv(dev);
+       u16 qid = skb_get_queue_mapping(skb);
 
        if (unlikely(pi->eth_flags & PRIV_FLAG_PORT_TX_VM))
                return cxgb4_vf_eth_xmit(skb, dev);
 
+       if (unlikely(qid >= pi->nqsets))
+               return cxgb4_ethofld_xmit(skb, dev);
+
        return cxgb4_eth_xmit(skb, dev);
 }
 
@@ -3311,6 +3365,22 @@ static int napi_rx_handler(struct napi_struct *napi, int budget)
        return work_done;
 }
 
+void cxgb4_ethofld_restart(unsigned long data)
+{
+       struct sge_eosw_txq *eosw_txq = (struct sge_eosw_txq *)data;
+       int pktcount;
+
+       spin_lock(&eosw_txq->lock);
+       pktcount = eosw_txq->cidx - eosw_txq->last_cidx;
+       if (pktcount < 0)
+               pktcount += eosw_txq->ndesc;
+
+       if (pktcount)
+               cxgb4_eosw_txq_free_desc(netdev2adap(eosw_txq->netdev),
+                                        eosw_txq, pktcount);
+       spin_unlock(&eosw_txq->lock);
+}
+
 /*
  * The MSI-X interrupt handler for an SGE response queue.
  */