Merge branch 'be2net-next'
Sathya Perla says:
====================
be2net: patch set
Hi David, this patch set includes 2 feature additions to the be2net driver:
Patch 1 sets up cpu affinity hints for be2net irqs using the
cpumask_set_cpu_local_first() API that first picks the near numa cores
and when they are exhausted, selects the far numa cores.
Patch 2 setups up xps queue mapping for be2net's TXQs to avoid,
by default, TX lock contention.
Patch 3 just bumps up the driver version.
Pls consider applying this patch set to the net-next queue. Thanks!
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ethernet/emulex/benet/be.h b/drivers/net/ethernet/emulex/benet/be.h
index eb39673..4b0494b 100644
--- a/drivers/net/ethernet/emulex/benet/be.h
+++ b/drivers/net/ethernet/emulex/benet/be.h
@@ -30,11 +30,12 @@
#include <linux/firmware.h>
#include <linux/slab.h>
#include <linux/u64_stats_sync.h>
+#include <linux/cpumask.h>
#include "be_hw.h"
#include "be_roce.h"
-#define DRV_VER "10.4u"
+#define DRV_VER "10.6.0.1"
#define DRV_NAME "be2net"
#define BE_NAME "Emulex BladeEngine2"
#define BE3_NAME "Emulex BladeEngine3"
@@ -183,6 +184,7 @@
u16 spurious_intr;
struct napi_struct napi;
struct be_adapter *adapter;
+ cpumask_var_t affinity_mask;
#ifdef CONFIG_NET_RX_BUSY_POLL
#define BE_EQ_IDLE 0
diff --git a/drivers/net/ethernet/emulex/benet/be_main.c b/drivers/net/ethernet/emulex/benet/be_main.c
index d8df78b..5ff7fba 100644
--- a/drivers/net/ethernet/emulex/benet/be_main.c
+++ b/drivers/net/ethernet/emulex/benet/be_main.c
@@ -2342,6 +2342,7 @@
napi_hash_del(&eqo->napi);
netif_napi_del(&eqo->napi);
}
+ free_cpumask_var(eqo->affinity_mask);
be_queue_free(adapter, &eqo->q);
}
}
@@ -2357,6 +2358,11 @@
adapter->cfg_num_qs);
for_all_evt_queues(adapter, eqo, i) {
+ if (!zalloc_cpumask_var(&eqo->affinity_mask, GFP_KERNEL))
+ return -ENOMEM;
+ cpumask_set_cpu_local_first(i, dev_to_node(&adapter->pdev->dev),
+ eqo->affinity_mask);
+
netif_napi_add(adapter->netdev, &eqo->napi, be_poll,
BE_NAPI_WEIGHT);
napi_hash_add(&eqo->napi);
@@ -2448,8 +2454,9 @@
static int be_tx_qs_create(struct be_adapter *adapter)
{
- struct be_queue_info *cq, *eq;
+ struct be_queue_info *cq;
struct be_tx_obj *txo;
+ struct be_eq_obj *eqo;
int status, i;
adapter->num_tx_qs = min(adapter->num_evt_qs, be_max_txqs(adapter));
@@ -2467,8 +2474,8 @@
/* If num_evt_qs is less than num_tx_qs, then more than
* one txq share an eq
*/
- eq = &adapter->eq_obj[i % adapter->num_evt_qs].q;
- status = be_cmd_cq_create(adapter, cq, eq, false, 3);
+ eqo = &adapter->eq_obj[i % adapter->num_evt_qs];
+ status = be_cmd_cq_create(adapter, cq, &eqo->q, false, 3);
if (status)
return status;
@@ -2480,6 +2487,9 @@
status = be_cmd_txq_create(adapter, txo);
if (status)
return status;
+
+ netif_set_xps_queue(adapter->netdev, eqo->affinity_mask,
+ eqo->idx);
}
dev_info(&adapter->pdev->dev, "created %d TX queue(s)\n",
@@ -3028,6 +3038,8 @@
status = request_irq(vec, be_msix, 0, eqo->desc, eqo);
if (status)
goto err_msix;
+
+ irq_set_affinity_hint(vec, eqo->affinity_mask);
}
return 0;
@@ -3072,7 +3084,7 @@
{
struct net_device *netdev = adapter->netdev;
struct be_eq_obj *eqo;
- int i;
+ int i, vec;
if (!adapter->isr_registered)
return;
@@ -3084,8 +3096,11 @@
}
/* MSIx */
- for_all_evt_queues(adapter, eqo, i)
- free_irq(be_msix_vec_get(adapter, eqo), eqo);
+ for_all_evt_queues(adapter, eqo, i) {
+ vec = be_msix_vec_get(adapter, eqo);
+ irq_set_affinity_hint(vec, NULL);
+ free_irq(vec, eqo);
+ }
done:
adapter->isr_registered = false;