[PATCH] e1000: Fix adapter structure and prepare for multique fix
Fix adapter structure to handle multiple queues and prepping the driver for full multiple queue support, some changes are ifdef'd our unless you define CONFIG_E1000_MQ.
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: John Ronciak <john.ronciak@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
diff --git a/drivers/net/e1000/e1000_main.c b/drivers/net/e1000/e1000_main.c
index 5f848b3..540c856 100644
--- a/drivers/net/e1000/e1000_main.c
+++ b/drivers/net/e1000/e1000_main.c
@@ -411,8 +411,9 @@
e1000_configure_tx(adapter);
e1000_setup_rctl(adapter);
e1000_configure_rx(adapter);
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_rx_queues; i++) {
adapter->alloc_rx_buf(adapter, &adapter->rx_ring[i]);
+ }
#ifdef CONFIG_PCI_MSI
if(adapter->hw.mac_type > e1000_82547_rev_2) {
@@ -867,7 +868,7 @@
unregister_netdev(netdev);
#ifdef CONFIG_E1000_NAPI
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_rx_queues; i++)
__dev_put(&adapter->polling_netdev[i]);
#endif
@@ -972,15 +973,25 @@
switch (hw->mac_type) {
case e1000_82571:
case e1000_82572:
- adapter->num_queues = 2;
+ /* These controllers support 2 tx queues, but with a single
+ * qdisc implementation, multiple tx queues aren't quite as
+ * interesting. If we can find a logical way of mapping
+ * flows to a queue, then perhaps we can up the num_tx_queue
+ * count back to its default. Until then, we run the risk of
+ * terrible performance due to SACK overload. */
+ adapter->num_tx_queues = 1;
+ adapter->num_rx_queues = 2;
break;
default:
- adapter->num_queues = 1;
+ adapter->num_tx_queues = 1;
+ adapter->num_rx_queues = 1;
break;
}
- adapter->num_queues = min(adapter->num_queues, num_online_cpus());
+ adapter->num_rx_queues = min(adapter->num_rx_queues, num_online_cpus());
+ adapter->num_tx_queues = min(adapter->num_tx_queues, num_online_cpus());
#else
- adapter->num_queues = 1;
+ adapter->num_tx_queues = 1;
+ adapter->num_rx_queues = 1;
#endif
if (e1000_alloc_queues(adapter)) {
@@ -989,7 +1000,7 @@
}
#ifdef CONFIG_E1000_NAPI
- for (i = 0; i < adapter->num_queues; i++) {
+ for (i = 0; i < adapter->num_rx_queues; i++) {
adapter->polling_netdev[i].priv = adapter;
adapter->polling_netdev[i].poll = &e1000_clean;
adapter->polling_netdev[i].weight = 64;
@@ -1022,13 +1033,13 @@
{
int size;
- size = sizeof(struct e1000_tx_ring) * adapter->num_queues;
+ size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
adapter->tx_ring = kmalloc(size, GFP_KERNEL);
if (!adapter->tx_ring)
return -ENOMEM;
memset(adapter->tx_ring, 0, size);
- size = sizeof(struct e1000_rx_ring) * adapter->num_queues;
+ size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
adapter->rx_ring = kmalloc(size, GFP_KERNEL);
if (!adapter->rx_ring) {
kfree(adapter->tx_ring);
@@ -1037,7 +1048,7 @@
memset(adapter->rx_ring, 0, size);
#ifdef CONFIG_E1000_NAPI
- size = sizeof(struct net_device) * adapter->num_queues;
+ size = sizeof(struct net_device) * adapter->num_rx_queues;
adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
if (!adapter->polling_netdev) {
kfree(adapter->tx_ring);
@@ -1066,12 +1077,12 @@
lock_cpu_hotplug();
i = 0;
for_each_online_cpu(cpu) {
- *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_queues];
+ *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_tx_queues];
/* This is incomplete because we'd like to assign separate
* physical cpus to these netdev polling structures and
* avoid saturating a subset of cpus.
*/
- if (i < adapter->num_queues) {
+ if (i < adapter->num_rx_queues) {
*per_cpu_ptr(adapter->cpu_netdev, cpu) = &adapter->polling_netdev[i];
adapter->cpu_for_queue[i] = cpu;
} else
@@ -1291,7 +1302,7 @@
{
int i, err = 0;
- for (i = 0; i < adapter->num_queues; i++) {
+ for (i = 0; i < adapter->num_tx_queues; i++) {
err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
if (err) {
DPRINTK(PROBE, ERR,
@@ -1319,7 +1330,7 @@
/* Setup the HW Tx Head and Tail descriptor pointers */
- switch (adapter->num_queues) {
+ switch (adapter->num_tx_queues) {
case 2:
tdba = adapter->tx_ring[1].dma;
tdlen = adapter->tx_ring[1].count *
@@ -1537,7 +1548,7 @@
{
int i, err = 0;
- for (i = 0; i < adapter->num_queues; i++) {
+ for (i = 0; i < adapter->num_rx_queues; i++) {
err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
if (err) {
DPRINTK(PROBE, ERR,
@@ -1709,7 +1720,7 @@
/* Setup the HW Rx Head and Tail Descriptor Pointers and
* the Base and Length of the Rx Descriptor Ring */
- switch (adapter->num_queues) {
+ switch (adapter->num_rx_queues) {
#ifdef CONFIG_E1000_MQ
case 2:
rdba = adapter->rx_ring[1].dma;
@@ -1736,7 +1747,7 @@
}
#ifdef CONFIG_E1000_MQ
- if (adapter->num_queues > 1) {
+ if (adapter->num_rx_queues > 1) {
uint32_t random[10];
get_random_bytes(&random[0], 40);
@@ -1746,7 +1757,7 @@
E1000_WRITE_REG(hw, RSSIM, 0);
}
- switch (adapter->num_queues) {
+ switch (adapter->num_rx_queues) {
case 2:
default:
reta = 0x00800080;
@@ -1838,7 +1849,7 @@
{
int i;
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_tx_queues; i++)
e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
}
@@ -1905,7 +1916,7 @@
{
int i;
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_tx_queues; i++)
e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
}
@@ -1949,7 +1960,7 @@
{
int i;
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_rx_queues; i++)
e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
}
@@ -2025,7 +2036,7 @@
{
int i;
- for (i = 0; i < adapter->num_queues; i++)
+ for (i = 0; i < adapter->num_rx_queues; i++)
e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
}
@@ -2325,7 +2336,10 @@
e1000_update_adaptive(&adapter->hw);
- if (adapter->num_queues == 1 && !netif_carrier_ok(netdev)) {
+#ifdef CONFIG_E1000_MQ
+ txdr = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
+#endif
+ if (!netif_carrier_ok(netdev)) {
if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
/* We've lost link, so the controller stops DMA,
* but we've got queued Tx work that's never going
@@ -3197,14 +3211,12 @@
E1000_WRITE_FLUSH(hw);
#ifdef CONFIG_E1000_MQ
if (atomic_read(&adapter->rx_sched_call_data.count) == 0) {
- cpu_set(adapter->cpu_for_queue[0],
- adapter->rx_sched_call_data.cpumask);
- for (i = 1; i < adapter->num_queues; i++) {
- cpu_set(adapter->cpu_for_queue[i],
- adapter->rx_sched_call_data.cpumask);
- atomic_inc(&adapter->irq_sem);
- }
- atomic_set(&adapter->rx_sched_call_data.count, i);
+ /* We must setup the cpumask once count == 0 since
+ * each cpu bit is cleared when the work is done. */
+ adapter->rx_sched_call_data.cpumask = adapter->cpumask;
+ atomic_add(adapter->num_rx_queues - 1, &adapter->irq_sem);
+ atomic_set(&adapter->rx_sched_call_data.count,
+ adapter->num_rx_queues);
smp_call_async_mask(&adapter->rx_sched_call_data);
} else {
printk("call_data.count == %u\n", atomic_read(&adapter->rx_sched_call_data.count));
@@ -3267,7 +3279,7 @@
while (poll_dev != &adapter->polling_netdev[i]) {
i++;
- if (unlikely(i == adapter->num_queues))
+ if (unlikely(i == adapter->num_rx_queues))
BUG();
}