enic: Add support for adaptive interrupt coalescing
This patch adds support for adaptive interrupt coalescing.
For small pkts with low pkt rate, we can decrease the coalescing interrupt
dynamically which decreases the latency. This however increases the cpu
utilization. Based on testing with different coal intr and pkt rate we came up
with a table(mod_table) with rx_rate and coalescing interrupt value where we
get low latency without significant increase in cpu. mod_table table stores
the coalescing timer percentage value for different throughputs.
Function enic_calc_int_moderation() calculates the desired coalescing intr timer
value. This function is called in driver rx napi_poll. The actual value is set
by enic_set_int_moderation() which is called when napi_poll is complete. i.e
when we unmask the rx intr.
Adaptive coal intr is support only when driver is using msix intr. Because
intr is not shared.
Struct mod_range is used to store only the default adaptive coalescing intr
value.
Adaptive coal intr calue is calculated by
timer = range_start + ((rx_coal->range_end - range_start) *
mod_table[index].range_percent / 100);
rx_coal->range_end is the rx-usecs-high value set using ethtool.
range_start is rx-usecs-low, set using ethtool, if rx_small_pkt_bytes_cnt is
greater than 2 * rx_large_pkt_bytes_cnt. i.e small pkts are dominant. Else its
rx-usecs-low + 3.
Cc: Christian Benvenuti <benve@cisco.com>
Cc: Neel Patel <neepatel@cisco.com>
Signed-off-by: Sujith Sankar <ssujith@cisco.com>
Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index 2945718..0d8995c 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/drivers/net/ethernet/cisco/enic/enic_main.c
@@ -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_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 @@
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 @@
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 @@
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 @@
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;
+
+ 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 @@
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 @@
*/
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_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 @@
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 @@
}
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))