| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1318347952-25068-6-git-send-email-jeffrey.t.kirsher@intel.com>
Date: Tue, 11 Oct 2011 08:45:46 -0700
From: Jeff Kirsher <jeffrey.t.kirsher@...el.com>
To: davem@...emloft.net
Cc: Alexander Duyck <alexander.h.duyck@...el.com>,
netdev@...r.kernel.org, gospo@...hat.com, sassmann@...hat.com,
Jeff Kirsher <jeffrey.t.kirsher@...el.com>
Subject: [net-next 05/11] igb: Move ITR related data into work container within the q_vector
From: Alexander Duyck <alexander.h.duyck@...el.com>
This change moves information related to interrupt throttle rate
configuration into a separate q_vector sub-structure called a work
container. A similar change has already been made for ixgbe and this work
is based off of that.
Signed-off-by: Alexander Duyck <alexander.h.duyck@...el.com>
Tested-by: Aaron Brown <aaron.f.brown@...el.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@...el.com>
---
drivers/net/ethernet/intel/igb/e1000_defines.h | 3 +
drivers/net/ethernet/intel/igb/igb.h | 32 +++--
drivers/net/ethernet/intel/igb/igb_ethtool.c | 4 +-
drivers/net/ethernet/intel/igb/igb_main.c | 203 +++++++++++-------------
4 files changed, 119 insertions(+), 123 deletions(-)
diff --git a/drivers/net/ethernet/intel/igb/e1000_defines.h b/drivers/net/ethernet/intel/igb/e1000_defines.h
index 7b8ddd8..68558be 100644
--- a/drivers/net/ethernet/intel/igb/e1000_defines.h
+++ b/drivers/net/ethernet/intel/igb/e1000_defines.h
@@ -409,6 +409,9 @@
#define E1000_ICS_DRSTA E1000_ICR_DRSTA /* Device Reset Aserted */
/* Extended Interrupt Cause Set */
+/* E1000_EITR_CNT_IGNR is only for 82576 and newer */
+#define E1000_EITR_CNT_IGNR 0x80000000 /* Don't reset counters on write */
+
/* Transmit Descriptor Control */
/* Enable the counting of descriptors still to be processed. */
diff --git a/drivers/net/ethernet/intel/igb/igb.h b/drivers/net/ethernet/intel/igb/igb.h
index 0cc4702..e0dea22 100644
--- a/drivers/net/ethernet/intel/igb/igb.h
+++ b/drivers/net/ethernet/intel/igb/igb.h
@@ -42,8 +42,11 @@
struct igb_adapter;
-/* ((1000000000ns / (6000ints/s * 1024ns)) << 2 = 648 */
-#define IGB_START_ITR 648
+/* Interrupt defines */
+#define IGB_START_ITR 648 /* ~6000 ints/sec */
+#define IGB_4K_ITR 980
+#define IGB_20K_ITR 196
+#define IGB_70K_ITR 56
/* TX/RX descriptor defines */
#define IGB_DEFAULT_TXD 256
@@ -175,15 +178,23 @@ struct igb_rx_queue_stats {
u64 alloc_failed;
};
+struct igb_ring_container {
+ struct igb_ring *ring; /* pointer to linked list of rings */
+ unsigned int total_bytes; /* total bytes processed this int */
+ unsigned int total_packets; /* total packets processed this int */
+ u16 work_limit; /* total work allowed per interrupt */
+ u8 count; /* total number of rings in vector */
+ u8 itr; /* current ITR setting for ring */
+};
+
struct igb_q_vector {
- struct igb_adapter *adapter; /* backlink */
- struct igb_ring *rx_ring;
- struct igb_ring *tx_ring;
- struct napi_struct napi;
+ struct igb_adapter *adapter; /* backlink */
+ int cpu; /* CPU for DCA */
+ u32 eims_value; /* EIMS mask value */
- u32 eims_value;
- u16 cpu;
- u16 tx_work_limit;
+ struct igb_ring_container rx, tx;
+
+ struct napi_struct napi;
u16 itr_val;
u8 set_itr;
@@ -213,9 +224,6 @@ struct igb_ring {
u16 next_to_clean ____cacheline_aligned_in_smp;
u16 next_to_use;
- unsigned int total_bytes;
- unsigned int total_packets;
-
union {
/* TX */
struct {
diff --git a/drivers/net/ethernet/intel/igb/igb_ethtool.c b/drivers/net/ethernet/intel/igb/igb_ethtool.c
index c364d84..0a9e309 100644
--- a/drivers/net/ethernet/intel/igb/igb_ethtool.c
+++ b/drivers/net/ethernet/intel/igb/igb_ethtool.c
@@ -2009,8 +2009,8 @@ static int igb_set_coalesce(struct net_device *netdev,
for (i = 0; i < adapter->num_q_vectors; i++) {
struct igb_q_vector *q_vector = adapter->q_vector[i];
- q_vector->tx_work_limit = adapter->tx_work_limit;
- if (q_vector->rx_ring)
+ q_vector->tx.work_limit = adapter->tx_work_limit;
+ if (q_vector->rx.ring)
q_vector->itr_val = adapter->rx_itr_setting;
else
q_vector->itr_val = adapter->tx_itr_setting;
diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c
index 91ae368..a5597de 100644
--- a/drivers/net/ethernet/intel/igb/igb_main.c
+++ b/drivers/net/ethernet/intel/igb/igb_main.c
@@ -737,10 +737,10 @@ static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
int rx_queue = IGB_N0_QUEUE;
int tx_queue = IGB_N0_QUEUE;
- if (q_vector->rx_ring)
- rx_queue = q_vector->rx_ring->reg_idx;
- if (q_vector->tx_ring)
- tx_queue = q_vector->tx_ring->reg_idx;
+ if (q_vector->rx.ring)
+ rx_queue = q_vector->rx.ring->reg_idx;
+ if (q_vector->tx.ring)
+ tx_queue = q_vector->tx.ring->reg_idx;
switch (hw->mac.type) {
case e1000_82575:
@@ -923,15 +923,15 @@ static int igb_request_msix(struct igb_adapter *adapter)
q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
- if (q_vector->rx_ring && q_vector->tx_ring)
+ if (q_vector->rx.ring && q_vector->tx.ring)
sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
- q_vector->rx_ring->queue_index);
- else if (q_vector->tx_ring)
+ q_vector->rx.ring->queue_index);
+ else if (q_vector->tx.ring)
sprintf(q_vector->name, "%s-tx-%u", netdev->name,
- q_vector->tx_ring->queue_index);
- else if (q_vector->rx_ring)
+ q_vector->tx.ring->queue_index);
+ else if (q_vector->rx.ring)
sprintf(q_vector->name, "%s-rx-%u", netdev->name,
- q_vector->rx_ring->queue_index);
+ q_vector->rx.ring->queue_index);
else
sprintf(q_vector->name, "%s-unused", netdev->name);
@@ -1110,8 +1110,9 @@ static void igb_map_rx_ring_to_vector(struct igb_adapter *adapter,
{
struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
- q_vector->rx_ring = adapter->rx_ring[ring_idx];
- q_vector->rx_ring->q_vector = q_vector;
+ q_vector->rx.ring = adapter->rx_ring[ring_idx];
+ q_vector->rx.ring->q_vector = q_vector;
+ q_vector->rx.count++;
q_vector->itr_val = adapter->rx_itr_setting;
if (q_vector->itr_val && q_vector->itr_val <= 3)
q_vector->itr_val = IGB_START_ITR;
@@ -1122,10 +1123,11 @@ static void igb_map_tx_ring_to_vector(struct igb_adapter *adapter,
{
struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
- q_vector->tx_ring = adapter->tx_ring[ring_idx];
- q_vector->tx_ring->q_vector = q_vector;
+ q_vector->tx.ring = adapter->tx_ring[ring_idx];
+ q_vector->tx.ring->q_vector = q_vector;
+ q_vector->tx.count++;
q_vector->itr_val = adapter->tx_itr_setting;
- q_vector->tx_work_limit = adapter->tx_work_limit;
+ q_vector->tx.work_limit = adapter->tx_work_limit;
if (q_vector->itr_val && q_vector->itr_val <= 3)
q_vector->itr_val = IGB_START_ITR;
}
@@ -3756,33 +3758,24 @@ static void igb_update_ring_itr(struct igb_q_vector *q_vector)
int new_val = q_vector->itr_val;
int avg_wire_size = 0;
struct igb_adapter *adapter = q_vector->adapter;
- struct igb_ring *ring;
unsigned int packets;
/* For non-gigabit speeds, just fix the interrupt rate at 4000
* ints/sec - ITR timer value of 120 ticks.
*/
if (adapter->link_speed != SPEED_1000) {
- new_val = 976;
+ new_val = IGB_4K_ITR;
goto set_itr_val;
}
- ring = q_vector->rx_ring;
- if (ring) {
- packets = ACCESS_ONCE(ring->total_packets);
-
- if (packets)
- avg_wire_size = ring->total_bytes / packets;
- }
+ packets = q_vector->rx.total_packets;
+ if (packets)
+ avg_wire_size = q_vector->rx.total_bytes / packets;
- ring = q_vector->tx_ring;
- if (ring) {
- packets = ACCESS_ONCE(ring->total_packets);
-
- if (packets)
- avg_wire_size = max_t(u32, avg_wire_size,
- ring->total_bytes / packets);
- }
+ packets = q_vector->tx.total_packets;
+ if (packets)
+ avg_wire_size = max_t(u32, avg_wire_size,
+ q_vector->tx.total_bytes / packets);
/* if avg_wire_size isn't set no work was done */
if (!avg_wire_size)
@@ -3800,9 +3793,11 @@ static void igb_update_ring_itr(struct igb_q_vector *q_vector)
else
new_val = avg_wire_size / 2;
- /* when in itr mode 3 do not exceed 20K ints/sec */
- if (adapter->rx_itr_setting == 3 && new_val < 196)
- new_val = 196;
+ /* conservative mode (itr 3) eliminates the lowest_latency setting */
+ if (new_val < IGB_20K_ITR &&
+ ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
+ (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
+ new_val = IGB_20K_ITR;
set_itr_val:
if (new_val != q_vector->itr_val) {
@@ -3810,14 +3805,10 @@ set_itr_val:
q_vector->set_itr = 1;
}
clear_counts:
- if (q_vector->rx_ring) {
- q_vector->rx_ring->total_bytes = 0;
- q_vector->rx_ring->total_packets = 0;
- }
- if (q_vector->tx_ring) {
- q_vector->tx_ring->total_bytes = 0;
- q_vector->tx_ring->total_packets = 0;
- }
+ q_vector->rx.total_bytes = 0;
+ q_vector->rx.total_packets = 0;
+ q_vector->tx.total_bytes = 0;
+ q_vector->tx.total_packets = 0;
}
/**
@@ -3833,106 +3824,102 @@ clear_counts:
* parameter (see igb_param.c)
* NOTE: These calculations are only valid when operating in a single-
* queue environment.
- * @adapter: pointer to adapter
- * @itr_setting: current q_vector->itr_val
- * @packets: the number of packets during this measurement interval
- * @bytes: the number of bytes during this measurement interval
+ * @q_vector: pointer to q_vector
+ * @ring_container: ring info to update the itr for
**/
-static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
- int packets, int bytes)
+static void igb_update_itr(struct igb_q_vector *q_vector,
+ struct igb_ring_container *ring_container)
{
- unsigned int retval = itr_setting;
+ unsigned int packets = ring_container->total_packets;
+ unsigned int bytes = ring_container->total_bytes;
+ u8 itrval = ring_container->itr;
+ /* no packets, exit with status unchanged */
if (packets == 0)
- goto update_itr_done;
+ return;
- switch (itr_setting) {
+ switch (itrval) {
case lowest_latency:
/* handle TSO and jumbo frames */
if (bytes/packets > 8000)
- retval = bulk_latency;
+ itrval = bulk_latency;
else if ((packets < 5) && (bytes > 512))
- retval = low_latency;
+ itrval = low_latency;
break;
case low_latency: /* 50 usec aka 20000 ints/s */
if (bytes > 10000) {
/* this if handles the TSO accounting */
if (bytes/packets > 8000) {
- retval = bulk_latency;
+ itrval = bulk_latency;
} else if ((packets < 10) || ((bytes/packets) > 1200)) {
- retval = bulk_latency;
+ itrval = bulk_latency;
} else if ((packets > 35)) {
- retval = lowest_latency;
+ itrval = lowest_latency;
}
} else if (bytes/packets > 2000) {
- retval = bulk_latency;
+ itrval = bulk_latency;
} else if (packets <= 2 && bytes < 512) {
- retval = lowest_latency;
+ itrval = lowest_latency;
}
break;
case bulk_latency: /* 250 usec aka 4000 ints/s */
if (bytes > 25000) {
if (packets > 35)
- retval = low_latency;
+ itrval = low_latency;
} else if (bytes < 1500) {
- retval = low_latency;
+ itrval = low_latency;
}
break;
}
-update_itr_done:
- return retval;
+ /* clear work counters since we have the values we need */
+ ring_container->total_bytes = 0;
+ ring_container->total_packets = 0;
+
+ /* write updated itr to ring container */
+ ring_container->itr = itrval;
}
-static void igb_set_itr(struct igb_adapter *adapter)
+static void igb_set_itr(struct igb_q_vector *q_vector)
{
- struct igb_q_vector *q_vector = adapter->q_vector[0];
- u16 current_itr;
+ struct igb_adapter *adapter = q_vector->adapter;
u32 new_itr = q_vector->itr_val;
+ u8 current_itr = 0;
/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
if (adapter->link_speed != SPEED_1000) {
current_itr = 0;
- new_itr = 4000;
+ new_itr = IGB_4K_ITR;
goto set_itr_now;
}
- adapter->rx_itr = igb_update_itr(adapter,
- adapter->rx_itr,
- q_vector->rx_ring->total_packets,
- q_vector->rx_ring->total_bytes);
+ igb_update_itr(q_vector, &q_vector->tx);
+ igb_update_itr(q_vector, &q_vector->rx);
- adapter->tx_itr = igb_update_itr(adapter,
- adapter->tx_itr,
- q_vector->tx_ring->total_packets,
- q_vector->tx_ring->total_bytes);
- current_itr = max(adapter->rx_itr, adapter->tx_itr);
+ current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
/* conservative mode (itr 3) eliminates the lowest_latency setting */
- if (adapter->rx_itr_setting == 3 && current_itr == lowest_latency)
+ if (current_itr == lowest_latency &&
+ ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
+ (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
current_itr = low_latency;
switch (current_itr) {
/* counts and packets in update_itr are dependent on these numbers */
case lowest_latency:
- new_itr = 56; /* aka 70,000 ints/sec */
+ new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
break;
case low_latency:
- new_itr = 196; /* aka 20,000 ints/sec */
+ new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
break;
case bulk_latency:
- new_itr = 980; /* aka 4,000 ints/sec */
+ new_itr = IGB_4K_ITR; /* 4,000 ints/sec */
break;
default:
break;
}
set_itr_now:
- q_vector->rx_ring->total_bytes = 0;
- q_vector->rx_ring->total_packets = 0;
- q_vector->tx_ring->total_bytes = 0;
- q_vector->tx_ring->total_packets = 0;
-
if (new_itr != q_vector->itr_val) {
/* this attempts to bias the interrupt rate towards Bulk
* by adding intermediate steps when interrupt rate is
@@ -3940,7 +3927,7 @@ set_itr_now:
new_itr = new_itr > q_vector->itr_val ?
max((new_itr * q_vector->itr_val) /
(new_itr + (q_vector->itr_val >> 2)),
- new_itr) :
+ new_itr) :
new_itr;
/* Don't write the value here; it resets the adapter's
* internal timer, and causes us to delay far longer than
@@ -4760,7 +4747,7 @@ static void igb_write_itr(struct igb_q_vector *q_vector)
if (adapter->hw.mac.type == e1000_82575)
itr_val |= itr_val << 16;
else
- itr_val |= 0x8000000;
+ itr_val |= E1000_EITR_CNT_IGNR;
writel(itr_val, q_vector->itr_register);
q_vector->set_itr = 0;
@@ -4788,8 +4775,8 @@ static void igb_update_dca(struct igb_q_vector *q_vector)
if (q_vector->cpu == cpu)
goto out_no_update;
- if (q_vector->tx_ring) {
- int q = q_vector->tx_ring->reg_idx;
+ if (q_vector->tx.ring) {
+ int q = q_vector->tx.ring->reg_idx;
u32 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
if (hw->mac.type == e1000_82575) {
dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
@@ -4802,8 +4789,8 @@ static void igb_update_dca(struct igb_q_vector *q_vector)
dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
}
- if (q_vector->rx_ring) {
- int q = q_vector->rx_ring->reg_idx;
+ if (q_vector->rx.ring) {
+ int q = q_vector->rx.ring->reg_idx;
u32 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
if (hw->mac.type == e1000_82575) {
dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
@@ -5447,16 +5434,14 @@ static irqreturn_t igb_intr(int irq, void *data)
/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
* need for the IMC write */
u32 icr = rd32(E1000_ICR);
- if (!icr)
- return IRQ_NONE; /* Not our interrupt */
-
- igb_write_itr(q_vector);
/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
* not set, then the adapter didn't send an interrupt */
if (!(icr & E1000_ICR_INT_ASSERTED))
return IRQ_NONE;
+ igb_write_itr(q_vector);
+
if (icr & E1000_ICR_DRSTA)
schedule_work(&adapter->reset_task);
@@ -5477,15 +5462,15 @@ static irqreturn_t igb_intr(int irq, void *data)
return IRQ_HANDLED;
}
-static inline void igb_ring_irq_enable(struct igb_q_vector *q_vector)
+void igb_ring_irq_enable(struct igb_q_vector *q_vector)
{
struct igb_adapter *adapter = q_vector->adapter;
struct e1000_hw *hw = &adapter->hw;
- if ((q_vector->rx_ring && (adapter->rx_itr_setting & 3)) ||
- (!q_vector->rx_ring && (adapter->tx_itr_setting & 3))) {
- if (!adapter->msix_entries)
- igb_set_itr(adapter);
+ if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
+ (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
+ if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
+ igb_set_itr(q_vector);
else
igb_update_ring_itr(q_vector);
}
@@ -5514,10 +5499,10 @@ static int igb_poll(struct napi_struct *napi, int budget)
if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_dca(q_vector);
#endif
- if (q_vector->tx_ring)
+ if (q_vector->tx.ring)
clean_complete = igb_clean_tx_irq(q_vector);
- if (q_vector->rx_ring)
+ if (q_vector->rx.ring)
clean_complete &= igb_clean_rx_irq(q_vector, budget);
/* If all work not completed, return budget and keep polling */
@@ -5597,11 +5582,11 @@ static void igb_tx_hwtstamp(struct igb_q_vector *q_vector,
static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
{
struct igb_adapter *adapter = q_vector->adapter;
- struct igb_ring *tx_ring = q_vector->tx_ring;
+ struct igb_ring *tx_ring = q_vector->tx.ring;
struct igb_tx_buffer *tx_buffer;
union e1000_adv_tx_desc *tx_desc, *eop_desc;
unsigned int total_bytes = 0, total_packets = 0;
- unsigned int budget = q_vector->tx_work_limit;
+ unsigned int budget = q_vector->tx.work_limit;
unsigned int i = tx_ring->next_to_clean;
if (test_bit(__IGB_DOWN, &adapter->state))
@@ -5687,8 +5672,8 @@ static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
tx_ring->tx_stats.bytes += total_bytes;
tx_ring->tx_stats.packets += total_packets;
u64_stats_update_end(&tx_ring->tx_syncp);
- tx_ring->total_bytes += total_bytes;
- tx_ring->total_packets += total_packets;
+ q_vector->tx.total_bytes += total_bytes;
+ q_vector->tx.total_packets += total_packets;
if (tx_ring->detect_tx_hung) {
struct e1000_hw *hw = &adapter->hw;
@@ -5837,7 +5822,7 @@ static inline u16 igb_get_hlen(union e1000_adv_rx_desc *rx_desc)
static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, int budget)
{
- struct igb_ring *rx_ring = q_vector->rx_ring;
+ struct igb_ring *rx_ring = q_vector->rx.ring;
union e1000_adv_rx_desc *rx_desc;
const int current_node = numa_node_id();
unsigned int total_bytes = 0, total_packets = 0;
@@ -5954,8 +5939,8 @@ next_desc:
rx_ring->rx_stats.packets += total_packets;
rx_ring->rx_stats.bytes += total_bytes;
u64_stats_update_end(&rx_ring->rx_syncp);
- rx_ring->total_packets += total_packets;
- rx_ring->total_bytes += total_bytes;
+ q_vector->rx.total_packets += total_packets;
+ q_vector->rx.total_bytes += total_bytes;
if (cleaned_count)
igb_alloc_rx_buffers(rx_ring, cleaned_count);
--
1.7.6.4
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists