IPoIB: Make send and receive queue sizes tunable

Make IPoIB's send and receive queue sizes tunable via module
parameters ("send_queue_size" and "recv_queue_size").  This allows the
queue sizes to be enlarged to fix disastrously bad performance on some
platforms and workloads, without bloating memory usage when large
queues aren't needed.

Signed-off-by: Shirley Ma <xma@us.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

4 files changed, 40 insertions, 20 deletions

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 374109df730..12a1e0572ef 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -65,6 +65,8 @@ enum {
 
         IPOIB_RX_RING_SIZE        = 128,
         IPOIB_TX_RING_SIZE        = 64,
+        IPOIB_MAX_QUEUE_SIZE      = 8192,
+        IPOIB_MIN_QUEUE_SIZE      = 2,
 
         IPOIB_NUM_WC              = 4,
 
@@ -332,6 +334,8 @@ static inline void ipoib_unregister_debugfs(void) { }
 #define ipoib_warn(priv, format, arg...)                \
         ipoib_printk(KERN_WARNING, priv, format , ## arg)
 
+extern int ipoib_sendq_size;
+extern int ipoib_recvq_size;
 
 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
 extern int ipoib_debug_level;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index ed65202878d..a54da42849a 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -161,7 +161,7 @@ static int ipoib_ib_post_receives(struct net_device *dev)
         struct ipoib_dev_priv *priv = netdev_priv(dev);
         int i;
 
-        for (i = 0; i < IPOIB_RX_RING_SIZE; ++i) {
+        for (i = 0; i < ipoib_recvq_size; ++i) {
                 if (ipoib_alloc_rx_skb(dev, i)) {
                         ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
                         return -ENOMEM;
@@ -187,7 +187,7 @@ static void ipoib_ib_handle_wc(struct net_device *dev,
         if (wr_id & IPOIB_OP_RECV) {
                 wr_id &= ~IPOIB_OP_RECV;
 
-                if (wr_id < IPOIB_RX_RING_SIZE) {
+                if (wr_id < ipoib_recvq_size) {
                         struct sk_buff *skb  = priv->rx_ring[wr_id].skb;
                         dma_addr_t      addr = priv->rx_ring[wr_id].mapping;
 
@@ -252,9 +252,9 @@ static void ipoib_ib_handle_wc(struct net_device *dev,
                 struct ipoib_tx_buf *tx_req;
                 unsigned long flags;
 
-                if (wr_id >= IPOIB_TX_RING_SIZE) {
+                if (wr_id >= ipoib_sendq_size) {
                         ipoib_warn(priv, "completion event with wrid %d (> %d)\n",
-                                   wr_id, IPOIB_TX_RING_SIZE);
+                                   wr_id, ipoib_sendq_size);
                         return;
                 }
 
@@ -275,7 +275,7 @@ static void ipoib_ib_handle_wc(struct net_device *dev,
                 spin_lock_irqsave(&priv->tx_lock, flags);
                 ++priv->tx_tail;
                 if (netif_queue_stopped(dev) &&
-                    priv->tx_head - priv->tx_tail <= IPOIB_TX_RING_SIZE / 2)
+                    priv->tx_head - priv->tx_tail <= ipoib_sendq_size >> 1)
                         netif_wake_queue(dev);
                 spin_unlock_irqrestore(&priv->tx_lock, flags);
 
@@ -344,13 +344,13 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
          * means we have to make sure everything is properly recorded and
          * our state is consistent before we call post_send().
          */
-        tx_req = &priv->tx_ring[priv->tx_head & (IPOIB_TX_RING_SIZE - 1)];
+        tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
         tx_req->skb = skb;
         addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
                               DMA_TO_DEVICE);
         pci_unmap_addr_set(tx_req, mapping, addr);
 
-        if (unlikely(post_send(priv, priv->tx_head & (IPOIB_TX_RING_SIZE - 1),
+        if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
                                address->ah, qpn, addr, skb->len))) {
                 ipoib_warn(priv, "post_send failed\n");
                 ++priv->stats.tx_errors;
@@ -363,7 +363,7 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
                 address->last_send = priv->tx_head;
                 ++priv->tx_head;
 
-                if (priv->tx_head - priv->tx_tail == IPOIB_TX_RING_SIZE) {
+                if (priv->tx_head - priv->tx_tail == ipoib_sendq_size) {
                         ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
                         netif_stop_queue(dev);
                 }
@@ -488,7 +488,7 @@ static int recvs_pending(struct net_device *dev)
         int pending = 0;
         int i;
 
-        for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
+        for (i = 0; i < ipoib_recvq_size; ++i)
                 if (priv->rx_ring[i].skb)
                         ++pending;
 
@@ -527,7 +527,7 @@ int ipoib_ib_dev_stop(struct net_device *dev)
                          */
                         while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
                                 tx_req = &priv->tx_ring[priv->tx_tail &
-                                                        (IPOIB_TX_RING_SIZE - 1)];
+                                                        (ipoib_sendq_size - 1)];
                                 dma_unmap_single(priv->ca->dma_device,
                                                  pci_unmap_addr(tx_req, mapping),
                                                  tx_req->skb->len,
@@ -536,7 +536,7 @@ int ipoib_ib_dev_stop(struct net_device *dev)
                                 ++priv->tx_tail;
                         }
 
-                        for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
+                        for (i = 0; i < ipoib_recvq_size; ++i)
                                 if (priv->rx_ring[i].skb) {
                                         dma_unmap_single(priv->ca->dma_device,
                                                          pci_unmap_addr(&priv->rx_ring[i],
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 9cb9e430aaa..5bf7e263454 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -41,6 +41,7 @@
 #include <linux/init.h>
 #include <linux/slab.h>
 #include <linux/vmalloc.h>
+#include <linux/kernel.h>
 
 #include <linux/if_arp.h>       /* For ARPHRD_xxx */
 
@@ -53,6 +54,14 @@ MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
 MODULE_LICENSE("Dual BSD/GPL");
 
+int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
+int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
+
+module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
+MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
+module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
+MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
+
 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
 int ipoib_debug_level;
 
@@ -795,20 +804,19 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
         struct ipoib_dev_priv *priv = netdev_priv(dev);
 
         /* Allocate RX/TX "rings" to hold queued skbs */
-
-        priv->rx_ring = kzalloc(IPOIB_RX_RING_SIZE * sizeof (struct ipoib_rx_buf),
+        priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
                                 GFP_KERNEL);
         if (!priv->rx_ring) {
                 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
-                       ca->name, IPOIB_RX_RING_SIZE);
+                       ca->name, ipoib_recvq_size);
                 goto out;
         }
 
-        priv->tx_ring = kzalloc(IPOIB_TX_RING_SIZE * sizeof (struct ipoib_tx_buf),
+        priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring,
                                 GFP_KERNEL);
         if (!priv->tx_ring) {
                 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
-                       ca->name, IPOIB_TX_RING_SIZE);
+                       ca->name, ipoib_sendq_size);
                 goto out_rx_ring_cleanup;
         }
 
@@ -876,7 +884,7 @@ static void ipoib_setup(struct net_device *dev)
         dev->hard_header_len     = IPOIB_ENCAP_LEN + INFINIBAND_ALEN;
         dev->addr_len            = INFINIBAND_ALEN;
         dev->type                = ARPHRD_INFINIBAND;
-        dev->tx_queue_len        = IPOIB_TX_RING_SIZE * 2;
+        dev->tx_queue_len        = ipoib_sendq_size * 2;
         dev->features            = NETIF_F_VLAN_CHALLENGED | NETIF_F_LLTX;
 
         /* MTU will be reset when mcast join happens */
@@ -1128,6 +1136,14 @@ static int __init ipoib_init_module(void)
 {
         int ret;
 
+        ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
+        ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
+        ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
+
+        ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
+        ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
+        ipoib_sendq_size = max(ipoib_sendq_size, IPOIB_MIN_QUEUE_SIZE);
+
         ret = ipoib_register_debugfs();
         if (ret)
                 return ret;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 5f0388027b2..1d49d1643c5 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -159,8 +159,8 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
         struct ipoib_dev_priv *priv = netdev_priv(dev);
         struct ib_qp_init_attr init_attr = {
                 .cap = {
-                        .max_send_wr  = IPOIB_TX_RING_SIZE,
-                        .max_recv_wr  = IPOIB_RX_RING_SIZE,
+                        .max_send_wr  = ipoib_sendq_size,
+                        .max_recv_wr  = ipoib_recvq_size,
                         .max_send_sge = 1,
                         .max_recv_sge = 1
                 },
@@ -175,7 +175,7 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
         }
 
         priv->cq = ib_create_cq(priv->ca, ipoib_ib_completion, NULL, dev,
-                                IPOIB_TX_RING_SIZE + IPOIB_RX_RING_SIZE + 1);
+                                ipoib_sendq_size + ipoib_recvq_size + 1);
         if (IS_ERR(priv->cq)) {
                 printk(KERN_WARNING "%s: failed to create CQ\n", ca->name);
                 goto out_free_pd;