enic: Fix 64 bit divide on 32bit system
[cascardo/linux.git] / drivers / net / ethernet / cisco / enic / enic_main.c
index 2945718..d5a220d 100644 (file)
@@ -38,6 +38,7 @@
 #include <linux/rtnetlink.h>
 #include <linux/prefetch.h>
 #include <net/ip6_checksum.h>
+#include <linux/ktime.h>
 
 #include "cq_enet_desc.h"
 #include "vnic_dev.h"
@@ -72,6 +73,35 @@ MODULE_LICENSE("GPL");
 MODULE_VERSION(DRV_VERSION);
 MODULE_DEVICE_TABLE(pci, enic_id_table);
 
+#define ENIC_LARGE_PKT_THRESHOLD               1000
+#define ENIC_MAX_COALESCE_TIMERS               10
+/*  Interrupt moderation table, which will be used to decide the
+ *  coalescing timer values
+ *  {rx_rate in Mbps, mapping percentage of the range}
+ */
+struct enic_intr_mod_table mod_table[ENIC_MAX_COALESCE_TIMERS + 1] = {
+       {4000,  0},
+       {4400, 10},
+       {5060, 20},
+       {5230, 30},
+       {5540, 40},
+       {5820, 50},
+       {6120, 60},
+       {6435, 70},
+       {6745, 80},
+       {7000, 90},
+       {0xFFFFFFFF, 100}
+};
+
+/* This table helps the driver to pick different ranges for rx coalescing
+ * timer depending on the link speed.
+ */
+struct enic_intr_mod_range mod_range[ENIC_MAX_LINK_SPEEDS] = {
+       {0,  0}, /* 0  - 4  Gbps */
+       {0,  3}, /* 4  - 10 Gbps */
+       {3,  6}, /* 10 - 40 Gbps */
+};
+
 int enic_is_dynamic(struct enic *enic)
 {
        return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
@@ -979,6 +1009,15 @@ static int enic_rq_alloc_buf(struct vnic_rq *rq)
        return 0;
 }
 
+static void enic_intr_update_pkt_size(struct vnic_rx_bytes_counter *pkt_size,
+                                     u32 pkt_len)
+{
+       if (ENIC_LARGE_PKT_THRESHOLD <= pkt_len)
+               pkt_size->large_pkt_bytes_cnt += pkt_len;
+       else
+               pkt_size->small_pkt_bytes_cnt += pkt_len;
+}
+
 static void enic_rq_indicate_buf(struct vnic_rq *rq,
        struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
        int skipped, void *opaque)
@@ -986,6 +1025,7 @@ static void enic_rq_indicate_buf(struct vnic_rq *rq,
        struct enic *enic = vnic_dev_priv(rq->vdev);
        struct net_device *netdev = enic->netdev;
        struct sk_buff *skb;
+       struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
 
        u8 type, color, eop, sop, ingress_port, vlan_stripped;
        u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
@@ -1056,6 +1096,9 @@ static void enic_rq_indicate_buf(struct vnic_rq *rq,
                        napi_gro_receive(&enic->napi[q_number], skb);
                else
                        netif_receive_skb(skb);
+               if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
+                       enic_intr_update_pkt_size(&cq->pkt_size_counter,
+                                                 bytes_written);
        } else {
 
                /* Buffer overflow
@@ -1134,6 +1177,64 @@ static int enic_poll(struct napi_struct *napi, int budget)
        return rq_work_done;
 }
 
+static void enic_set_int_moderation(struct enic *enic, struct vnic_rq *rq)
+{
+       unsigned int intr = enic_msix_rq_intr(enic, rq->index);
+       struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
+       u32 timer = cq->tobe_rx_coal_timeval;
+
+       if (cq->tobe_rx_coal_timeval != cq->cur_rx_coal_timeval) {
+               vnic_intr_coalescing_timer_set(&enic->intr[intr], timer);
+               cq->cur_rx_coal_timeval = cq->tobe_rx_coal_timeval;
+       }
+}
+
+static void enic_calc_int_moderation(struct enic *enic, struct vnic_rq *rq)
+{
+       struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
+       struct vnic_cq *cq = &enic->cq[enic_cq_rq(enic, rq->index)];
+       struct vnic_rx_bytes_counter *pkt_size_counter = &cq->pkt_size_counter;
+       int index;
+       u32 timer;
+       u32 range_start;
+       u32 traffic;
+       u64 delta;
+       ktime_t now = ktime_get();
+
+       delta = ktime_us_delta(now, cq->prev_ts);
+       if (delta < ENIC_AIC_TS_BREAK)
+               return;
+       cq->prev_ts = now;
+
+       traffic = pkt_size_counter->large_pkt_bytes_cnt +
+                 pkt_size_counter->small_pkt_bytes_cnt;
+       /* The table takes Mbps
+        * traffic *= 8    => bits
+        * traffic *= (10^6 / delta)    => bps
+        * traffic /= 10^6     => Mbps
+        *
+        * Combining, traffic *= (8 / delta)
+        */
+
+       traffic <<= 3;
+       traffic = delta > UINT_MAX ? 0 : traffic / (u32)delta;
+
+       for (index = 0; index < ENIC_MAX_COALESCE_TIMERS; index++)
+               if (traffic < mod_table[index].rx_rate)
+                       break;
+       range_start = (pkt_size_counter->small_pkt_bytes_cnt >
+                      pkt_size_counter->large_pkt_bytes_cnt << 1) ?
+                     rx_coal->small_pkt_range_start :
+                     rx_coal->large_pkt_range_start;
+       timer = range_start + ((rx_coal->range_end - range_start) *
+                              mod_table[index].range_percent / 100);
+       /* Damping */
+       cq->tobe_rx_coal_timeval = (timer + cq->tobe_rx_coal_timeval) >> 1;
+
+       pkt_size_counter->large_pkt_bytes_cnt = 0;
+       pkt_size_counter->small_pkt_bytes_cnt = 0;
+}
+
 static int enic_poll_msix(struct napi_struct *napi, int budget)
 {
        struct net_device *netdev = napi->dev;
@@ -1171,6 +1272,13 @@ static int enic_poll_msix(struct napi_struct *napi, int budget)
 
        if (err)
                work_done = work_to_do;
+       if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
+               /* Call the function which refreshes
+                * the intr coalescing timer value based on
+                * the traffic.  This is supported only in
+                * the case of MSI-x mode
+                */
+               enic_calc_int_moderation(enic, &enic->rq[rq]);
 
        if (work_done < work_to_do) {
 
@@ -1179,6 +1287,8 @@ static int enic_poll_msix(struct napi_struct *napi, int budget)
                 */
 
                napi_complete(napi);
+               if (enic->rx_coalesce_setting.use_adaptive_rx_coalesce)
+                       enic_set_int_moderation(enic, &enic->rq[rq]);
                vnic_intr_unmask(&enic->intr[intr]);
        }
 
@@ -1314,6 +1424,42 @@ static void enic_synchronize_irqs(struct enic *enic)
        }
 }
 
+static void enic_set_rx_coal_setting(struct enic *enic)
+{
+       unsigned int speed;
+       int index = -1;
+       struct enic_rx_coal *rx_coal = &enic->rx_coalesce_setting;
+
+       /* If intr mode is not MSIX, do not do adaptive coalescing */
+       if (VNIC_DEV_INTR_MODE_MSIX != vnic_dev_get_intr_mode(enic->vdev)) {
+               netdev_info(enic->netdev, "INTR mode is not MSIX, Not initializing adaptive coalescing");
+               return;
+       }
+
+       /* 1. Read the link speed from fw
+        * 2. Pick the default range for the speed
+        * 3. Update it in enic->rx_coalesce_setting
+        */
+       speed = vnic_dev_port_speed(enic->vdev);
+       if (ENIC_LINK_SPEED_10G < speed)
+               index = ENIC_LINK_40G_INDEX;
+       else if (ENIC_LINK_SPEED_4G < speed)
+               index = ENIC_LINK_10G_INDEX;
+       else
+               index = ENIC_LINK_4G_INDEX;
+
+       rx_coal->small_pkt_range_start = mod_range[index].small_pkt_range_start;
+       rx_coal->large_pkt_range_start = mod_range[index].large_pkt_range_start;
+       rx_coal->range_end = ENIC_RX_COALESCE_RANGE_END;
+
+       /* Start with the value provided by UCSM */
+       for (index = 0; index < enic->rq_count; index++)
+               enic->cq[index].cur_rx_coal_timeval =
+                               enic->config.intr_timer_usec;
+
+       rx_coal->use_adaptive_rx_coalesce = 1;
+}
+
 static int enic_dev_notify_set(struct enic *enic)
 {
        int err;
@@ -2231,6 +2377,7 @@ static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        enic->notify_timer.function = enic_notify_timer;
        enic->notify_timer.data = (unsigned long)enic;
 
+       enic_set_rx_coal_setting(enic);
        INIT_WORK(&enic->reset, enic_reset);
        INIT_WORK(&enic->change_mtu_work, enic_change_mtu_work);
 
@@ -2250,6 +2397,9 @@ static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        }
 
        enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
+       /* rx coalesce time already got initialized. This gets used
+        * if adaptive coal is turned off
+        */
        enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
 
        if (enic_is_dynamic(enic) || enic_is_sriov_vf(enic))