* [PATCH] make tg3 NAPI support configurable
@ 2003-12-19 12:48 Greg Banks
2003-12-19 13:20 ` Robert Olsson
0 siblings, 1 reply; 7+ messages in thread
From: Greg Banks @ 2003-12-19 12:48 UTC (permalink / raw)
To: David S. Miller, Jeff Garzik; +Cc: Linux Network Development list
G'day,
I've been having some issues with irq rates and cpu usage in the
tg3 driver. In short, on Altix machines they're far too high.
It turned out that reverting the driver to its pre-NAPI interrupt
coalescing scheme made the situation a lot better.
How much better? Running 8192 byte UDP packets across gige
with NAPI takes 99.5% of a CPU to service 29,100 irqs per second.
With the pre-NAPI code the figures are 36.0% CPU and 4880 irq/sec.
Similar improvements are seen for non-fragmented UDP and for TCP.
So, this patch against 2.6.0-test11 makes NAPI support in the tg3
driver selectable with a config option. The default value is to
enable NAPI, on the assumption that NAPI must be working for most
people. The non-NAPI case is the code from before NAPI was added,
with tweaks for locking and ethtool_ops.
The patch also applies against 2.6.0 with a few offsets.
===========================================================================
linux/drivers/net/Kconfig
===========================================================================
--- /usr/tmp/TmpDir.7858-0/linux/drivers/net/Kconfig_1.22 Thu Dec 18 11:45:52 2003
+++ linux/drivers/net/Kconfig Thu Dec 18 11:45:11 2003
@@ -2017,6 +2017,14 @@
To compile this driver as a module, choose M here: the module
will be called tg3. This is recommended.
+config TIGON3_NAPI
+ bool "Use Rx Polling (NAPI)"
+ default y
+ depends on TIGON3
+ help
+ NAPI is a new driver API designed to reduce CPU and interrupt load
+ when the driver is receiving lots of packets from the card.
+
endmenu
#
===========================================================================
linux/drivers/net/tg3.c
===========================================================================
--- /usr/tmp/TmpDir.26498-0/linux/drivers/net/tg3.c_1.20 Tue Nov 11 12:38:57 2003
+++ linux/drivers/net/tg3.c Tue Nov 11 12:03:11 2003
@@ -2306,7 +2306,13 @@
desc->err_vlan & RXD_VLAN_MASK);
} else
#endif
+ {
+#ifdef CONFIG_TIGON3_NAPI
netif_receive_skb(skb);
+#else /* !CONFIG_TIGON3_NAPI */
+ netif_rx(skb);
+#endif /* !CONFIG_TIGON3_NAPI */
+ }
tp->dev->last_rx = jiffies;
received++;
@@ -2345,6 +2351,146 @@
return received;
}
+#ifndef CONFIG_TIGON3_NAPI
+
+#define PKT_RATE_LOW 22000
+#define PKT_RATE_HIGH 61000
+
+static void tg3_rate_sample(struct tg3 *tp, unsigned long ticks)
+{
+ u32 delta, rx_now, tx_now;
+ int new_vals, do_tx, do_rx;
+
+ rx_now = tp->hw_stats->rx_ucast_packets.low;
+ tx_now = tp->hw_stats->COS_out_packets[0].low;
+
+ delta = (rx_now - tp->last_rx_count);
+ delta += (tx_now - tp->last_tx_count);
+ delta /= (ticks / tp->coalesce_config.rate_sample_jiffies);
+
+ tp->last_rx_count = rx_now;
+ tp->last_tx_count = tx_now;
+
+ new_vals = 0;
+ do_tx = (tp->tg3_flags2 & TG3_FLG2_ADAPTIVE_TX) != 0;
+ do_rx = (tp->tg3_flags2 & TG3_FLG2_ADAPTIVE_RX) != 0;
+ if (delta < tp->coalesce_config.pkt_rate_low) {
+ if (do_rx &&
+ tp->coalesce_config.rx_max_coalesced_frames !=
+ tp->coalesce_config.rx_max_coalesced_frames_low) {
+ tp->coalesce_config.rx_max_coalesced_frames =
+ LOW_RXMAX_FRAMES;
+ tp->coalesce_config.rx_coalesce_ticks =
+ LOW_RXCOL_TICKS;
+ new_vals = 1;
+ }
+ if (do_tx &&
+ tp->coalesce_config.tx_max_coalesced_frames !=
+ tp->coalesce_config.tx_max_coalesced_frames_low) {
+ tp->coalesce_config.tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_low;
+ tp->coalesce_config.tx_coalesce_ticks =
+ tp->coalesce_config.tx_coalesce_ticks_low;
+ new_vals = 1;
+ }
+ } else if (delta < tp->coalesce_config.pkt_rate_high) {
+ if (do_rx &&
+ tp->coalesce_config.rx_max_coalesced_frames !=
+ tp->coalesce_config.rx_max_coalesced_frames_def) {
+ tp->coalesce_config.rx_max_coalesced_frames =
+ tp->coalesce_config.rx_max_coalesced_frames_def;
+ tp->coalesce_config.rx_coalesce_ticks =
+ tp->coalesce_config.rx_coalesce_ticks_def;
+ new_vals = 1;
+ }
+ if (do_tx &&
+ tp->coalesce_config.tx_max_coalesced_frames !=
+ tp->coalesce_config.tx_max_coalesced_frames_def) {
+ tp->coalesce_config.tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_def;
+ tp->coalesce_config.tx_coalesce_ticks =
+ tp->coalesce_config.tx_coalesce_ticks_def;
+ new_vals = 1;
+ }
+ } else {
+ if (do_rx &&
+ tp->coalesce_config.rx_max_coalesced_frames !=
+ tp->coalesce_config.rx_max_coalesced_frames_high) {
+ tp->coalesce_config.rx_max_coalesced_frames =
+ tp->coalesce_config.rx_max_coalesced_frames_high;
+ tp->coalesce_config.rx_coalesce_ticks =
+ tp->coalesce_config.rx_coalesce_ticks_high;
+ new_vals = 1;
+ }
+ if (do_tx &&
+ tp->coalesce_config.tx_max_coalesced_frames !=
+ tp->coalesce_config.tx_max_coalesced_frames_high) {
+ tp->coalesce_config.tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_high;
+ tp->coalesce_config.tx_coalesce_ticks =
+ tp->coalesce_config.tx_coalesce_ticks_high;
+ new_vals = 1;
+ }
+ }
+
+ if (new_vals) {
+ if (do_rx) {
+ tw32(HOSTCC_RXCOL_TICKS,
+ tp->coalesce_config.rx_coalesce_ticks);
+ tw32(HOSTCC_RXMAX_FRAMES,
+ tp->coalesce_config.rx_max_coalesced_frames);
+ }
+ if (do_tx) {
+ tw32(HOSTCC_TXCOL_TICKS,
+ tp->coalesce_config.tx_coalesce_ticks);
+ tw32(HOSTCC_TXMAX_FRAMES,
+ tp->coalesce_config.tx_max_coalesced_frames);
+ }
+ }
+
+ tp->last_rate_sample = jiffies;
+}
+
+static void tg3_interrupt_main_work(struct tg3 *tp)
+{
+ struct tg3_hw_status *sblk = tp->hw_status;
+ int did_pkts;
+
+ if (!(tp->tg3_flags &
+ (TG3_FLAG_USE_LINKCHG_REG |
+ TG3_FLAG_POLL_SERDES))) {
+ if (sblk->status & SD_STATUS_LINK_CHG) {
+ sblk->status = SD_STATUS_UPDATED |
+ (sblk->status & ~SD_STATUS_LINK_CHG);
+ tg3_setup_phy(tp);
+ }
+ }
+
+ /* gnbTODO: check locking here */
+
+ did_pkts = 0;
+ if (sblk->idx[0].rx_producer != tp->rx_rcb_ptr) {
+ tg3_rx(tp, 1024);
+ did_pkts = 1;
+ }
+
+ if (sblk->idx[0].tx_consumer != tp->tx_cons) {
+ spin_lock(&tp->tx_lock);
+ tg3_tx(tp);
+ spin_unlock(&tp->tx_lock);
+ }
+
+ if (did_pkts &&
+ (tp->tg3_flags2 & (TG3_FLG2_ADAPTIVE_RX | TG3_FLG2_ADAPTIVE_TX))) {
+ unsigned long ticks = jiffies - tp->last_rate_sample;
+
+ if (ticks >= tp->coalesce_config.rate_sample_jiffies)
+ tg3_rate_sample(tp, ticks);
+ }
+}
+
+#else /* CONFIG_TIGON3_NAPI */
+
static int tg3_poll(struct net_device *netdev, int *budget)
{
struct tg3 *tp = netdev->priv;
@@ -2406,6 +2552,8 @@
return (done ? 0 : 1);
}
+#endif /* CONFIG_TIGON3_NAPI */
+
static inline unsigned int tg3_has_work(struct net_device *dev, struct tg3 *tp)
{
struct tg3_hw_status *sblk = tp->hw_status;
@@ -2432,10 +2580,12 @@
struct tg3 *tp = dev->priv;
struct tg3_hw_status *sblk = tp->hw_status;
unsigned long flags;
- unsigned int handled = 1;
+ unsigned int handled;
spin_lock_irqsave(&tp->lock, flags);
+#ifdef CONFIG_TIGON3_NAPI
+ handled = 1;
if (sblk->status & SD_STATUS_UPDATED) {
/*
* writing any value to intr-mbox-0 clears PCI INTA# and
@@ -2466,6 +2616,21 @@
} else { /* shared interrupt */
handled = 0;
}
+#else
+ handled = 0;
+ while (sblk->status & SD_STATUS_UPDATED) {
+ handled = 1;
+ tw32_mailbox(MAILBOX_INTERRUPT_0 + TG3_64BIT_REG_LOW,
+ 0x00000001);
+ sblk->status &= ~SD_STATUS_UPDATED;
+
+ tg3_interrupt_main_work(tp);
+
+ tw32_mailbox(MAILBOX_INTERRUPT_0 + TG3_64BIT_REG_LOW,
+ 0x00000000);
+ tr32(MAILBOX_INTERRUPT_0 + TG3_64BIT_REG_LOW);
+ }
+#endif
spin_unlock_irqrestore(&tp->lock, flags);
@@ -4894,6 +5059,7 @@
udelay(10);
}
+#ifdef CONFIG_TIGON3_NAPI
tw32(HOSTCC_RXCOL_TICKS, 0);
tw32(HOSTCC_TXCOL_TICKS, LOW_TXCOL_TICKS);
tw32(HOSTCC_RXMAX_FRAMES, 1);
@@ -4904,6 +5070,28 @@
tw32(HOSTCC_TXCOAL_TICK_INT, 0);
tw32(HOSTCC_RXCOAL_MAXF_INT, 1);
tw32(HOSTCC_TXCOAL_MAXF_INT, 0);
+#else /* !CONFIG_TIGON3_NAPI */
+ tw32(HOSTCC_RXCOL_TICKS,
+ tp->coalesce_config.rx_coalesce_ticks);
+ tw32(HOSTCC_RXMAX_FRAMES,
+ tp->coalesce_config.rx_max_coalesced_frames);
+ if (GET_ASIC_REV(tp->pci_chip_rev_id) != ASIC_REV_5705)
+ tw32(HOSTCC_RXCOAL_TICK_INT,
+ tp->coalesce_config.rx_coalesce_ticks_during_int);
+ tw32(HOSTCC_RXCOAL_MAXF_INT,
+ tp->coalesce_config.rx_max_coalesced_frames_during_int);
+ tw32(HOSTCC_TXCOL_TICKS,
+ tp->coalesce_config.tx_coalesce_ticks);
+ tw32(HOSTCC_TXMAX_FRAMES,
+ tp->coalesce_config.tx_max_coalesced_frames);
+ if (GET_ASIC_REV(tp->pci_chip_rev_id) != ASIC_REV_5705)
+ tw32(HOSTCC_TXCOAL_TICK_INT,
+ tp->coalesce_config.tx_coalesce_ticks_during_int);
+ tw32(HOSTCC_TXCOAL_MAXF_INT,
+ tp->coalesce_config.tx_max_coalesced_frames_during_int);
+ tw32(HOSTCC_STAT_COAL_TICKS,
+ tp->coalesce_config.stats_coalesce_ticks);
+#endif /* !CONFIG_TIGON3_NAPI */
/* set status block DMA address */
tw32(HOSTCC_STATUS_BLK_HOST_ADDR + TG3_64BIT_REG_HIGH,
@@ -5922,8 +6110,13 @@
cmd->phy_address = PHY_ADDR;
cmd->transceiver = 0;
cmd->autoneg = tp->link_config.autoneg;
+#ifdef CONFIG_TIGON3_NAPI
cmd->maxtxpkt = 0;
cmd->maxrxpkt = 0;
+#else
+ cmd->maxtxpkt = tp->coalesce_config.tx_max_coalesced_frames_def;
+ cmd->maxrxpkt = tp->coalesce_config.rx_max_coalesced_frames_def;
+#endif
return 0;
}
@@ -5944,6 +6137,25 @@
tp->link_config.duplex = cmd->duplex;
}
+#ifndef CONFIG_TIGON3_NAPI
+ if (cmd->maxtxpkt || cmd->maxrxpkt) {
+ tp->coalesce_config.tx_max_coalesced_frames_def =
+ tp->coalesce_config.tx_max_coalesced_frames =
+ cmd->maxtxpkt;
+ tp->coalesce_config.rx_max_coalesced_frames_def =
+ tp->coalesce_config.rx_max_coalesced_frames =
+ cmd->maxrxpkt;
+
+ /* Coalescing config bits can be updated without
+ * a full chip reset.
+ */
+ tw32(HOSTCC_TXMAX_FRAMES,
+ tp->coalesce_config.tx_max_coalesced_frames);
+ tw32(HOSTCC_RXMAX_FRAMES,
+ tp->coalesce_config.rx_max_coalesced_frames);
+ }
+#endif /* !CONFIG_TIGON3_NAPI */
+
tg3_setup_phy(tp);
spin_unlock(&tp->tx_lock);
spin_unlock_irq(&tp->lock);
@@ -6036,7 +6248,186 @@
return r;
}
-
+
+#ifndef CONFIG_TIGON3_NAPI
+static int tg3_get_coalesce(struct net_device *dev,
+ struct ethtool_coalesce *ecoal)
+{
+ struct tg3 *tp = dev->priv;
+
+ ecoal->rx_coalesce_usecs =
+ tp->coalesce_config.rx_coalesce_ticks_def;
+ ecoal->rx_max_coalesced_frames =
+ tp->coalesce_config.rx_max_coalesced_frames_def;
+ ecoal->rx_coalesce_usecs_irq =
+ tp->coalesce_config.rx_coalesce_ticks_during_int_def;
+ ecoal->rx_max_coalesced_frames_irq =
+ tp->coalesce_config.rx_max_coalesced_frames_during_int_def;
+
+ ecoal->tx_coalesce_usecs =
+ tp->coalesce_config.tx_coalesce_ticks_def;
+ ecoal->tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_def;
+ ecoal->tx_coalesce_usecs_irq =
+ tp->coalesce_config.tx_coalesce_ticks_during_int_def;
+ ecoal->tx_max_coalesced_frames_irq =
+ tp->coalesce_config.tx_max_coalesced_frames_during_int_def;
+
+ ecoal->stats_block_coalesce_usecs =
+ tp->coalesce_config.stats_coalesce_ticks_def;
+
+ ecoal->use_adaptive_rx_coalesce =
+ (tp->tg3_flags2 & TG3_FLG2_ADAPTIVE_RX) != 0;
+ ecoal->use_adaptive_tx_coalesce =
+ (tp->tg3_flags2 & TG3_FLG2_ADAPTIVE_TX) != 0;
+
+ ecoal->pkt_rate_low =
+ tp->coalesce_config.pkt_rate_low;
+ ecoal->rx_coalesce_usecs_low =
+ tp->coalesce_config.rx_coalesce_ticks_low;
+ ecoal->rx_max_coalesced_frames_low =
+ tp->coalesce_config.rx_max_coalesced_frames_low;
+ ecoal->tx_coalesce_usecs_low =
+ tp->coalesce_config.tx_coalesce_ticks_low;
+ ecoal->tx_max_coalesced_frames_low =
+ tp->coalesce_config.tx_max_coalesced_frames_low;
+
+ ecoal->pkt_rate_high =
+ tp->coalesce_config.pkt_rate_high;
+ ecoal->rx_coalesce_usecs_high =
+ tp->coalesce_config.rx_coalesce_ticks_high;
+ ecoal->rx_max_coalesced_frames_high =
+ tp->coalesce_config.rx_max_coalesced_frames_high;
+ ecoal->tx_coalesce_usecs_high =
+ tp->coalesce_config.tx_coalesce_ticks_high;
+ ecoal->tx_max_coalesced_frames_high =
+ tp->coalesce_config.tx_max_coalesced_frames_high;
+
+ ecoal->rate_sample_interval =
+ tp->coalesce_config.rate_sample_jiffies / HZ;
+
+ return 0;
+}
+
+static int tg3_set_coalesce(struct net_device *dev,
+ struct ethtool_coalesce *ecoal)
+{
+ struct tg3 *tp = dev->priv;
+
+ /* Make sure we are not getting garbage. */
+ if ((ecoal->rx_coalesce_usecs == 0 &&
+ ecoal->rx_max_coalesced_frames == 0) ||
+ (ecoal->tx_coalesce_usecs == 0 &&
+ ecoal->tx_max_coalesced_frames == 0) ||
+ ecoal->stats_block_coalesce_usecs == 0)
+ return -EINVAL;
+ if (ecoal->use_adaptive_rx_coalesce ||
+ ecoal->use_adaptive_tx_coalesce) {
+ if (ecoal->pkt_rate_low > ecoal->pkt_rate_high)
+ return -EINVAL;
+ if (ecoal->rate_sample_interval == 0)
+ return -EINVAL;
+ if (ecoal->use_adaptive_rx_coalesce &&
+ ((ecoal->rx_coalesce_usecs_low == 0 &&
+ ecoal->rx_max_coalesced_frames_low == 0) ||
+ (ecoal->rx_coalesce_usecs_high == 0 &&
+ ecoal->rx_max_coalesced_frames_high == 0)))
+ return -EINVAL;
+ if (ecoal->use_adaptive_tx_coalesce &&
+ ((ecoal->tx_coalesce_usecs_low == 0 &&
+ ecoal->tx_max_coalesced_frames_low == 0) ||
+ (ecoal->tx_coalesce_usecs_high == 0 &&
+ ecoal->tx_max_coalesced_frames_high == 0)))
+ return -EINVAL;
+ }
+
+ /* Looks good, let it rip. */
+ spin_lock_irq(&tp->lock);
+ tp->coalesce_config.rx_coalesce_ticks =
+ tp->coalesce_config.rx_coalesce_ticks_def =
+ ecoal->rx_coalesce_usecs;
+ tp->coalesce_config.rx_max_coalesced_frames =
+ tp->coalesce_config.rx_max_coalesced_frames_def =
+ ecoal->rx_max_coalesced_frames;
+ tp->coalesce_config.rx_coalesce_ticks_during_int =
+ tp->coalesce_config.rx_coalesce_ticks_during_int_def =
+ ecoal->rx_coalesce_usecs_irq;
+ tp->coalesce_config.rx_max_coalesced_frames_during_int =
+ tp->coalesce_config.rx_max_coalesced_frames_during_int_def =
+ ecoal->rx_max_coalesced_frames_irq;
+ tp->coalesce_config.tx_coalesce_ticks =
+ tp->coalesce_config.tx_coalesce_ticks_def =
+ ecoal->tx_coalesce_usecs;
+ tp->coalesce_config.tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_def =
+ ecoal->tx_max_coalesced_frames;
+ tp->coalesce_config.tx_coalesce_ticks_during_int =
+ tp->coalesce_config.tx_coalesce_ticks_during_int_def =
+ ecoal->tx_coalesce_usecs_irq;
+ tp->coalesce_config.tx_max_coalesced_frames_during_int =
+ tp->coalesce_config.tx_max_coalesced_frames_during_int_def =
+ ecoal->tx_max_coalesced_frames_irq;
+ tp->coalesce_config.stats_coalesce_ticks =
+ tp->coalesce_config.stats_coalesce_ticks_def =
+ ecoal->stats_block_coalesce_usecs;
+
+ if (ecoal->use_adaptive_rx_coalesce)
+ tp->tg3_flags2 |= TG3_FLG2_ADAPTIVE_RX;
+ else
+ tp->tg3_flags2 &= ~TG3_FLG2_ADAPTIVE_RX;
+ if (ecoal->use_adaptive_tx_coalesce)
+ tp->tg3_flags2 |= TG3_FLG2_ADAPTIVE_TX;
+ else
+ tp->tg3_flags2 &= ~TG3_FLG2_ADAPTIVE_TX;
+
+ tp->coalesce_config.pkt_rate_low = ecoal->pkt_rate_low;
+ tp->coalesce_config.pkt_rate_high = ecoal->pkt_rate_high;
+ tp->coalesce_config.rate_sample_jiffies =
+ ecoal->rate_sample_interval * HZ;
+
+ tp->coalesce_config.rx_coalesce_ticks_low =
+ ecoal->rx_coalesce_usecs_low;
+ tp->coalesce_config.rx_max_coalesced_frames_low =
+ ecoal->rx_max_coalesced_frames_low;
+ tp->coalesce_config.tx_coalesce_ticks_low =
+ ecoal->tx_coalesce_usecs_low;
+ tp->coalesce_config.tx_max_coalesced_frames_low =
+ ecoal->tx_max_coalesced_frames_low;
+
+ tp->coalesce_config.rx_coalesce_ticks_high =
+ ecoal->rx_coalesce_usecs_high;
+ tp->coalesce_config.rx_max_coalesced_frames_high =
+ ecoal->rx_max_coalesced_frames_high;
+ tp->coalesce_config.tx_coalesce_ticks_high =
+ ecoal->tx_coalesce_usecs_high;
+ tp->coalesce_config.tx_max_coalesced_frames_high =
+ ecoal->tx_max_coalesced_frames_high;
+
+ tw32(HOSTCC_RXCOL_TICKS,
+ tp->coalesce_config.rx_coalesce_ticks_def);
+ tw32(HOSTCC_RXMAX_FRAMES,
+ tp->coalesce_config.rx_max_coalesced_frames_def);
+ tw32(HOSTCC_RXCOAL_TICK_INT,
+ tp->coalesce_config.rx_coalesce_ticks_during_int_def);
+ tw32(HOSTCC_RXCOAL_MAXF_INT,
+ tp->coalesce_config.rx_max_coalesced_frames_during_int_def);
+ tw32(HOSTCC_TXCOL_TICKS,
+ tp->coalesce_config.tx_coalesce_ticks_def);
+ tw32(HOSTCC_TXMAX_FRAMES,
+ tp->coalesce_config.tx_max_coalesced_frames_def);
+ tw32(HOSTCC_TXCOAL_TICK_INT,
+ tp->coalesce_config.tx_coalesce_ticks_during_int_def);
+ tw32(HOSTCC_TXCOAL_MAXF_INT,
+ tp->coalesce_config.tx_max_coalesced_frames_during_int_def);
+ tw32(HOSTCC_STAT_COAL_TICKS,
+ tp->coalesce_config.stats_coalesce_ticks_def);
+
+ spin_unlock_irq(&tp->lock);
+
+ return 0;
+}
+#endif /* !CONFIG_TIGON3_NAPI */
+
static void tg3_get_ringparam(struct net_device *dev, struct ethtool_ringparam
*ering)
{
struct tg3 *tp = dev->priv;
@@ -6245,6 +6636,10 @@
.set_msglevel = tg3_set_msglevel,
.nway_reset = tg3_nway_reset,
.get_link = ethtool_op_get_link,
+#ifndef CONFIG_TIGON3_NAPI
+ .get_coalesce = tg3_get_coalesce,
+ .set_coalesce = tg3_set_coalesce,
+#endif
.get_ringparam = tg3_get_ringparam,
.set_ringparam = tg3_set_ringparam,
.get_pauseparam = tg3_get_pauseparam,
@@ -6889,6 +7284,7 @@
(tp->pci_chip_rev_id != CHIPREV_ID_5705_A1)))
tp->tg3_flags2 |= TG3_FLG2_NO_ETH_WIRE_SPEED;
+#ifdef CONFIG_TIGON3_NAPI
/* Only 5701 and later support tagged irq status mode.
* Also, 5788 chips cannot use tagged irq status.
*
@@ -6897,6 +7293,10 @@
* fully clear in that mode.
*/
tp->coalesce_mode = 0;
+#else
+ tp->coalesce_mode = (HOSTCC_MODE_CLRTICK_RXBD |
+ HOSTCC_MODE_CLRTICK_TXBD);
+#endif
if (GET_CHIP_REV(tp->pci_chip_rev_id) != CHIPREV_5700_AX &&
GET_CHIP_REV(tp->pci_chip_rev_id) != CHIPREV_5700_BX)
@@ -7424,6 +7824,63 @@
tp->link_config.orig_autoneg = AUTONEG_INVALID;
}
+#ifndef CONFIG_TIGON3_NAPI
+static void __devinit tg3_init_coalesce_config(struct tg3 *tp)
+{
+ tp->coalesce_config.rx_coalesce_ticks_def = DEFAULT_RXCOL_TICKS;
+ tp->coalesce_config.rx_max_coalesced_frames_def = DEFAULT_RXMAX_FRAMES;
+ tp->coalesce_config.rx_coalesce_ticks_during_int_def =
+ DEFAULT_RXCOAL_TICK_INT;
+ tp->coalesce_config.rx_max_coalesced_frames_during_int_def =
+ DEFAULT_RXCOAL_MAXF_INT;
+ tp->coalesce_config.tx_coalesce_ticks_def = DEFAULT_TXCOL_TICKS;
+ tp->coalesce_config.tx_max_coalesced_frames_def = DEFAULT_TXMAX_FRAMES;
+ tp->coalesce_config.tx_coalesce_ticks_during_int_def =
+ DEFAULT_TXCOAL_TICK_INT;
+ tp->coalesce_config.tx_max_coalesced_frames_during_int_def =
+ DEFAULT_TXCOAL_MAXF_INT;
+ tp->coalesce_config.stats_coalesce_ticks_def =
+ DEFAULT_STAT_COAL_TICKS;
+
+ tp->coalesce_config.rx_coalesce_ticks_low =
+ LOW_RXCOL_TICKS;
+ tp->coalesce_config.rx_max_coalesced_frames_low =
+ LOW_RXMAX_FRAMES;
+ tp->coalesce_config.tx_coalesce_ticks_low =
+ LOW_TXCOL_TICKS;
+ tp->coalesce_config.tx_max_coalesced_frames_low =
+ LOW_TXMAX_FRAMES;
+
+ tp->coalesce_config.rx_coalesce_ticks_high =
+ HIGH_RXCOL_TICKS;
+ tp->coalesce_config.rx_max_coalesced_frames_high =
+ HIGH_RXMAX_FRAMES;
+ tp->coalesce_config.tx_coalesce_ticks_high =
+ HIGH_TXCOL_TICKS;
+ tp->coalesce_config.tx_max_coalesced_frames_high =
+ HIGH_TXMAX_FRAMES;
+
+ /* Active == default */
+ tp->coalesce_config.rx_coalesce_ticks =
+ tp->coalesce_config.rx_coalesce_ticks_def;
+ tp->coalesce_config.rx_max_coalesced_frames =
+ tp->coalesce_config.rx_max_coalesced_frames_def;
+ tp->coalesce_config.tx_coalesce_ticks =
+ tp->coalesce_config.tx_coalesce_ticks_def;
+ tp->coalesce_config.tx_max_coalesced_frames =
+ tp->coalesce_config.tx_max_coalesced_frames_def;
+ tp->coalesce_config.stats_coalesce_ticks =
+ tp->coalesce_config.stats_coalesce_ticks_def;
+
+ tp->coalesce_config.rate_sample_jiffies = (1 * HZ);
+ tp->coalesce_config.pkt_rate_low = 22000;
+ tp->coalesce_config.pkt_rate_high = 61000;
+
+ tp->tg3_flags2 |= TG3_FLG2_ADAPTIVE_RX;
+ tp->tg3_flags2 &= ~(TG3_FLG2_ADAPTIVE_TX);
+}
+#endif /* !CONFIG_TIGON3_NAPI */
+
static void __devinit tg3_init_bufmgr_config(struct tg3 *tp)
{
tp->bufmgr_config.mbuf_read_dma_low_water =
@@ -7613,6 +8070,10 @@
tg3_init_link_config(tp);
+#ifndef CONFIG_TIGON3_NAPI
+ tg3_init_coalesce_config(tp);
+#endif /* !CONFIG_TIGON3_NAPI */
+
tg3_init_bufmgr_config(tp);
tp->rx_pending = TG3_DEF_RX_RING_PENDING;
@@ -7626,9 +8087,11 @@
dev->set_mac_address = tg3_set_mac_addr;
dev->do_ioctl = tg3_ioctl;
dev->tx_timeout = tg3_tx_timeout;
- dev->poll = tg3_poll;
dev->ethtool_ops = &tg3_ethtool_ops;
+#ifdef CONFIG_TIGON3_NAPI
+ dev->poll = tg3_poll;
dev->weight = 64;
+#endif /* CONFIG_TIGON3_NAPI */
dev->watchdog_timeo = TG3_TX_TIMEOUT;
dev->change_mtu = tg3_change_mtu;
dev->irq = pdev->irq;
@@ -7718,7 +8181,7 @@
*/
pci_save_state(tp->pdev, tp->pci_cfg_state);
- printk(KERN_INFO "%s: Tigon3 [partno(%s) rev %04x PHY(%s)] (PCI%s:%s:%s) %sBaseT
Ethernet ",
+ printk(KERN_INFO "%s: Tigon3 [partno(%s) rev %04x PHY(%s)] (PCI%s:%s:%s) %sBaseT
Ethernet (%s) ",
dev->name,
tp->board_part_number,
tp->pci_chip_rev_id,
@@ -7728,7 +8191,13 @@
((tp->tg3_flags & TG3_FLAG_PCIX_MODE) ? "133MHz" : "66MHz") :
((tp->tg3_flags & TG3_FLAG_PCIX_MODE) ? "100MHz" : "33MHz")),
((tp->tg3_flags & TG3_FLAG_PCI_32BIT) ? "32-bit" : "64-bit"),
- (tp->tg3_flags & TG3_FLAG_10_100_ONLY) ? "10/100" : "10/100/1000");
+ (tp->tg3_flags & TG3_FLAG_10_100_ONLY) ? "10/100" : "10/100/1000",
+#ifdef CONFIG_TIGON3_NAPI
+ "NAPI"
+#else
+ "no NAPI"
+#endif
+ );
for (i = 0; i < 6; i++)
printk("%2.2x%c", dev->dev_addr[i],
===========================================================================
linux/drivers/net/tg3.h
===========================================================================
--- /usr/tmp/TmpDir.26498-0/linux/drivers/net/tg3.h_1.9 Tue Nov 11 12:38:57 2003
+++ linux/drivers/net/tg3.h Tue Nov 11 12:38:24 2003
@@ -1801,6 +1801,47 @@
u8 orig_autoneg;
};
+#ifndef CONFIG_TIGON3_NAPI
+struct tg3_coalesce_config {
+ /* Current settings. */
+ u32 rx_coalesce_ticks;
+ u32 rx_max_coalesced_frames;
+ u32 rx_coalesce_ticks_during_int;
+ u32 rx_max_coalesced_frames_during_int;
+ u32 tx_coalesce_ticks;
+ u32 tx_max_coalesced_frames;
+ u32 tx_coalesce_ticks_during_int;
+ u32 tx_max_coalesced_frames_during_int;
+ u32 stats_coalesce_ticks;
+
+ /* Default settings. */
+ u32 rx_coalesce_ticks_def;
+ u32 rx_max_coalesced_frames_def;
+ u32 rx_coalesce_ticks_during_int_def;
+ u32 rx_max_coalesced_frames_during_int_def;
+ u32 tx_coalesce_ticks_def;
+ u32 tx_max_coalesced_frames_def;
+ u32 tx_coalesce_ticks_during_int_def;
+ u32 tx_max_coalesced_frames_during_int_def;
+ u32 stats_coalesce_ticks_def;
+
+ /* Adaptive RX/TX coalescing parameters. */
+ u32 rate_sample_jiffies;
+ u32 pkt_rate_low;
+ u32 pkt_rate_high;
+
+ u32 rx_coalesce_ticks_low;
+ u32 rx_max_coalesced_frames_low;
+ u32 tx_coalesce_ticks_low;
+ u32 tx_max_coalesced_frames_low;
+
+ u32 rx_coalesce_ticks_high;
+ u32 rx_max_coalesced_frames_high;
+ u32 tx_coalesce_ticks_high;
+ u32 tx_max_coalesced_frames_high;
+};
+#endif /* !CONFIG_TIGON3_NAPI */
+
struct tg3_bufmgr_config {
u32 mbuf_read_dma_low_water;
u32 mbuf_mac_rx_low_water;
@@ -1879,6 +1920,13 @@
struct net_device_stats net_stats_prev;
unsigned long phy_crc_errors;
+#ifndef CONFIG_TIGON3_NAPI
+ /* Adaptive coalescing engine. */
+ unsigned long last_rate_sample;
+ u32 last_rx_count;
+ u32 last_tx_count;
+#endif /* !CONFIG_TIGON3_NAPI */
+
u32 rx_offset;
u32 tg3_flags;
#define TG3_FLAG_HOST_TXDS 0x00000001
@@ -1920,6 +1968,8 @@
#define TG3_FLG2_IS_5788 0x00000008
#define TG3_FLG2_MAX_RXPEND_64 0x00000010
#define TG3_FLG2_TSO_CAPABLE 0x00000020
+#define TG3_FLG2_ADAPTIVE_RX 0x00000040
+#define TG3_FLG2_ADAPTIVE_TX 0x00000080
u32 split_mode_max_reqs;
#define SPLIT_MODE_5704_MAX_REQ 3
@@ -1932,6 +1982,9 @@
u16 asf_multiplier;
struct tg3_link_config link_config;
+#ifndef CONFIG_TIGON3_NAPI
+ struct tg3_coalesce_config coalesce_config;
+#endif /* !CONFIG_TIGON3_NAPI */
struct tg3_bufmgr_config bufmgr_config;
/* cache h/w values, often passed straight to h/w */
Greg.
--
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
I don't speak for SGI.
^ permalink raw reply [flat|nested] 7+ messages in thread* [PATCH] make tg3 NAPI support configurable
2003-12-19 12:48 [PATCH] make tg3 NAPI support configurable Greg Banks
@ 2003-12-19 13:20 ` Robert Olsson
2004-01-11 1:49 ` Jeff Garzik
0 siblings, 1 reply; 7+ messages in thread
From: Robert Olsson @ 2003-12-19 13:20 UTC (permalink / raw)
To: Greg Banks; +Cc: David S. Miller, Jeff Garzik, Linux Network Development list
Greg Banks writes:
> I've been having some issues with irq rates and cpu usage in the
> tg3 driver. In short, on Altix machines they're far too high.
> It turned out that reverting the driver to its pre-NAPI interrupt
> coalescing scheme made the situation a lot better.
>
> How much better? Running 8192 byte UDP packets across gige
> with NAPI takes 99.5% of a CPU to service 29,100 irqs per second.
> With the pre-NAPI code the figures are 36.0% CPU and 4880 irq/sec.
> Similar improvements are seen for non-fragmented UDP and for TCP.
Hello!
You can use coalescing with NAPI as well, e1000 and other drivers
are doing this. This will give you same interrupt rates as non-
NAPI at low load and "polling" without any interrupts at high load.
Furthermore NAPI can be extended to schedule dev->poll even for TX-
interrupts. There is pacth for e1000 doing this. We see about 5-8%
overall system packet improvement with this.
Cheers.
--ro
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] make tg3 NAPI support configurable
2003-12-19 13:20 ` Robert Olsson
@ 2004-01-11 1:49 ` Jeff Garzik
2004-01-12 0:12 ` Greg Banks
2004-01-13 19:09 ` Robert Olsson
0 siblings, 2 replies; 7+ messages in thread
From: Jeff Garzik @ 2004-01-11 1:49 UTC (permalink / raw)
To: Robert Olsson; +Cc: Greg Banks, David S. Miller, Linux Network Development list
Robert Olsson wrote:
> Greg Banks writes:
>
> > I've been having some issues with irq rates and cpu usage in the
> > tg3 driver. In short, on Altix machines they're far too high.
> > It turned out that reverting the driver to its pre-NAPI interrupt
> > coalescing scheme made the situation a lot better.
> >
> > How much better? Running 8192 byte UDP packets across gige
> > with NAPI takes 99.5% of a CPU to service 29,100 irqs per second.
> > With the pre-NAPI code the figures are 36.0% CPU and 4880 irq/sec.
> > Similar improvements are seen for non-fragmented UDP and for TCP.
>
> Hello!
>
> You can use coalescing with NAPI as well, e1000 and other drivers
> are doing this. This will give you same interrupt rates as non-
> NAPI at low load and "polling" without any interrupts at high load.
Yes, this is something I've been meaning to add to tg3 for months now.
Adding some about of hardware intr mitigation -in addition to- NAPI will
not only help on the NAPI "hard case" of moderate load and a super-fast
CPU, but also help avoid certain silicon bugs...
> Furthermore NAPI can be extended to schedule dev->poll even for TX-
> interrupts. There is pacth for e1000 doing this. We see about 5-8%
> overall system packet improvement with this.
tg3 already schedules for TX, so we've got that part covered :)
Jeff
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] make tg3 NAPI support configurable
2004-01-11 1:49 ` Jeff Garzik
@ 2004-01-12 0:12 ` Greg Banks
2004-01-13 19:09 ` Robert Olsson
1 sibling, 0 replies; 7+ messages in thread
From: Greg Banks @ 2004-01-12 0:12 UTC (permalink / raw)
To: Jeff Garzik
Cc: Robert Olsson, David S. Miller, Linux Network Development list
Jeff Garzik wrote:
>
> Robert Olsson wrote:
> > You can use coalescing with NAPI as well, e1000 and other drivers
> > are doing this. This will give you same interrupt rates as non-
> > NAPI at low load and "polling" without any interrupts at high load.
>
> Yes, this is something I've been meaning to add to tg3 for months now.
I'd be very happy to test patches on an Altix, where this issue is a
significant networking scaling limitation.
Greg.
--
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
I don't speak for SGI.
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] make tg3 NAPI support configurable
2004-01-11 1:49 ` Jeff Garzik
2004-01-12 0:12 ` Greg Banks
@ 2004-01-13 19:09 ` Robert Olsson
1 sibling, 0 replies; 7+ messages in thread
From: Robert Olsson @ 2004-01-13 19:09 UTC (permalink / raw)
To: Jeff Garzik
Cc: Robert Olsson, Greg Banks, David S. Miller,
Linux Network Development list, jchapman
Jeff Garzik writes:
> > Furthermore NAPI can be extended to schedule dev->poll even for TX-
> > interrupts. There is pacth for e1000 doing this. We see about 5-8%
> > overall system packet improvement with this.
>
> tg3 already schedules for TX, so we've got that part covered :)
Hello!
I was thinking of a variant JC [jchapman@katalix.com] mentioned on this list
some time ago. He also sent me the patch for e1000. A test and the patch is
below.
Routing test.
============
2 * 10 Million pkts @ 2*783 kpps into eth0, eth2 routed to eth1, eth3.
(TX-OK is the number to look for)
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flags
eth0 1500 0 3494625 8258316 8258316 6505378 24 0 0 0 BRU
eth1 1500 0 45 0 0 0 3494627 0 0 0 BRU
eth2 1500 0 3493930 8270692 8270692 6506073 21 0 0 0 BRU
eth3 1500 0 1 0 0 0 3493929 0 0 0 BRU
CPU0
26: 74 IO-APIC-level eth0
27: 48617 IO-APIC-level eth1
28: 71 IO-APIC-level eth2
29: 48659 IO-APIC-level eth3
-------------------------------------------------------------------------------
With patch.
Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flags
eth0 1500 0 3752858 8151787 8151787 6247146 23 0 0 0 BRU
eth1 1500 0 47 0 0 0 3751676 0 0 0 BRU
eth2 1500 0 3751226 8191511 8191511 6248777 21 0 0 0 BRU
eth3 1500 0 1 0 0 0 3750490 0 0 0 BRU
CPU0
26: 125 IO-APIC-level eth0
27: 127 IO-APIC-level eth1
28: 122 IO-APIC-level eth2
29: 137 IO-APIC-level eth3
TX interrupts alone now schedules consecutive polls. We route 7.5 Million
pkts w/o any interrupts. Total throughput from 0.349% to 0.375% (~580 kpps)
Of course having having RX-only and TX-only is special case...
TCP-stream test.
================
Netperf w. single TCP-stream recv showed 938 Mbit/s both with and without
patch and interrupts rates were the same. XEON @ 2.66 GHz w. e1000 4-port
board. Linux 2.6.0-test11/UP
--- e1000_main.c.orig 2003-08-26 22:59:00.000000000 +0100
+++ e1000_main.c 2003-08-26 23:03:35.000000000 +0100
@@ -2061,19 +2061,21 @@
struct e1000_adapter *adapter = netdev->priv;
int work_to_do = min(*budget, netdev->quota);
int work_done = 0;
-
- e1000_clean_tx_irq(adapter);
+ boolean_t tx_cleaned;
+
+ tx_cleaned = e1000_clean_tx_irq(adapter);
e1000_clean_rx_irq(adapter, &work_done, work_to_do);
- *budget -= work_done;
- netdev->quota -= work_done;
-
- if(work_done < work_to_do) {
+ if(!tx_cleaned && (work_done == 0)) {
netif_rx_complete(netdev);
e1000_irq_enable(adapter);
+ return 0;
}
- return (work_done >= work_to_do);
+ *budget -= work_done;
+ netdev->quota -= work_done;
+
+ return 1;
}
#endif
Cheers.
--ro
^ permalink raw reply [flat|nested] 7+ messages in thread
* RE: [PATCH] make tg3 NAPI support configurable
@ 2004-01-14 0:23 Feldman, Scott
0 siblings, 0 replies; 7+ messages in thread
From: Feldman, Scott @ 2004-01-14 0:23 UTC (permalink / raw)
To: Robert Olsson, Jeff Garzik
Cc: Greg Banks, David S. Miller, Linux Network Development list,
jchapman
> I was thinking of a variant JC [jchapman@katalix.com]
> mentioned on this list some time ago. He also sent me
> the patch for e1000. A test and the patch is below.
JC contributed almost the exact patch for the e100 rewrite and it did
help Tx, but I don't remember how much. JC, do you remember? Here is
the snippet:
static int e100_poll(struct net_device *netdev, int *budget)
{
struct nic *nic = netdev->priv;
unsigned int work_to_do = min(netdev->quota, *budget);
unsigned int work_done = 0;
int tx_cleaned;
e100_rx_clean(nic, &work_done, work_to_do);
tx_cleaned = e100_tx_clean(nic);
/* If no Rx and Tx cleanup work was done, exit polling mode. */
if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev))
{
netif_rx_complete(netdev);
e100_enable_irq(nic);
return 0;
}
*budget -= work_done;
netdev->quota -= work_done;
return 1;
}
^ permalink raw reply [flat|nested] 7+ messages in thread* RE: [PATCH] make tg3 NAPI support configurable
@ 2004-01-15 8:01 jc
0 siblings, 0 replies; 7+ messages in thread
From: jc @ 2004-01-15 8:01 UTC (permalink / raw)
To: scott.feldman, Robert.Olsson, jgarzik; +Cc: gnb, davem, netdev
Hi all,
I found that testing if any tx work is done in dev->poll before exiting
polled mode improves performance by about 7% (max) in a 2-port e100 bridge
forwarding unidirectional test case. If tx work is not considered when
deciding whether to netif_rx_complete, the transmitting interface sees
loads of interrupts and hence forwarding throughput is degraded.
When testing with bidirectional test data, no improvement is seen since
the dev->poll is kept in polled mode on both interfaces due to
receive work.
Hope this helps.
-jc
> > > Furthermore NAPI can be extended to schedule dev->poll even for TX-
> > > interrupts. There is pacth for e1000 doing this. We see about 5-8%
> > > overall system packet improvement with this.
> > I was thinking of a variant JC [jchapman@katalix.com]
> > mentioned on this list some time ago. He also sent me
> > the patch for e1000. A test and the patch is below.
>
> JC contributed almost the exact patch for the e100 rewrite and it did
> help Tx, but I don't remember how much. JC, do you remember? Here is
> the snippet:
>
> static int e100_poll(struct net_device *netdev, int *budget)
> {
> struct nic *nic = netdev->priv;
> unsigned int work_to_do = min(netdev->quota, *budget);
> unsigned int work_done = 0;
> int tx_cleaned;
>
> e100_rx_clean(nic, &work_done, work_to_do);
> tx_cleaned = e100_tx_clean(nic);
>
> /* If no Rx and Tx cleanup work was done, exit polling mode. */
> if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev))
> {
> netif_rx_complete(netdev);
> e100_enable_irq(nic);
> return 0;
> }
>
> *budget -= work_done;
> netdev->quota -= work_done;
>
> return 1;
> }
-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2004-01-15 8:01 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-12-19 12:48 [PATCH] make tg3 NAPI support configurable Greg Banks
2003-12-19 13:20 ` Robert Olsson
2004-01-11 1:49 ` Jeff Garzik
2004-01-12 0:12 ` Greg Banks
2004-01-13 19:09 ` Robert Olsson
-- strict thread matches above, loose matches on Subject: below --
2004-01-14 0:23 Feldman, Scott
2004-01-15 8:01 jc
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).