Message-Id: <1205809191-20901-4-git-send-email-dhananjay@netxen.com>
Date:	Mon, 17 Mar 2008 19:59:50 -0700
From:	Dhananjay Phadke <dhananjay@...xen.com>
To:	netdev@...r.kernel.org
Cc:	jeff@...zik.org
Subject: [PATCH 3/4] [RESEND] netxen: remove low level tx lock

o eliminate tx lock in netxen adapter struct, instead pound on netdev
  tx lock appropriately.
o remove old "concurrent transmit" code that unnecessarily drops and
  reacquires tx lock in hard_xmit_frame(), this is already serialized
  the netdev xmit lock.
o reduce scope of tx lock in tx cleanup. tx cleanup operates on
  different section of the ring than transmitting cpus and is
  guarded by producer and consumer indices. This fixes a race
  caused by rx softirq preemption on realtime kernels.

Signed-off-by: Dhananjay Phadke <dhananjay@...xen.com>
Tested-by: Vernon Mauery <mauery@...ibm.com>
---
 drivers/net/netxen/netxen_nic.h         |   14 +----
 drivers/net/netxen/netxen_nic_ethtool.c |    2 -
 drivers/net/netxen/netxen_nic_init.c    |   89 ++++++-----------------------
 drivers/net/netxen/netxen_nic_main.c    |   95 +++++++-----------------------
 4 files changed, 43 insertions(+), 157 deletions(-)

diff --git a/drivers/net/netxen/netxen_nic.h b/drivers/net/netxen/netxen_nic.h
index 8b6546c..070421b 100644
--- a/drivers/net/netxen/netxen_nic.h
+++ b/drivers/net/netxen/netxen_nic.h
@@ -85,7 +85,7 @@
 	(sizeof(struct netxen_cmd_buffer) * adapter->max_tx_desc_count)
 #define RCV_BUFFSIZE	\
 	(sizeof(struct netxen_rx_buffer) * rcv_desc->max_rx_desc_count)
-#define find_diff_among(a,b,range) ((a)<=(b)?((b)-(a)):((b)+(range)-(a)))
+#define find_diff_among(a,b,range) ((a)<(b)?((b)-(a)):((b)+(range)-(a)))
 
 #define NETXEN_NETDEV_STATUS		0x1
 #define NETXEN_RCV_PRODUCER_OFFSET	0
@@ -204,7 +204,7 @@ enum {
 			? RCV_DESC_LRO :	\
 			(RCV_DESC_NORMAL)))
 
-#define MAX_CMD_DESCRIPTORS		1024
+#define MAX_CMD_DESCRIPTORS		4096
 #define MAX_RCV_DESCRIPTORS		16384
 #define MAX_CMD_DESCRIPTORS_HOST	(MAX_CMD_DESCRIPTORS / 4)
 #define MAX_RCV_DESCRIPTORS_1G		(MAX_RCV_DESCRIPTORS / 4)
@@ -824,9 +824,7 @@ struct netxen_adapter_stats {
 	u64  uphcong;
 	u64  upmcong;
 	u64  updunno;
-	u64  skbfreed;
 	u64  txdropped;
-	u64  txnullskb;
 	u64  csummed;
 	u64  no_rcv;
 	u64  rxbytes;
@@ -888,8 +886,6 @@ struct netxen_adapter {
 	int mtu;
 	int portnum;
 
-	spinlock_t tx_lock;
-	spinlock_t lock;
 	struct work_struct watchdog_task;
 	struct timer_list watchdog_timer;
 	struct work_struct  tx_timeout_task;
@@ -898,16 +894,12 @@ struct netxen_adapter {
 
 	u32 cmd_producer;
 	__le32 *cmd_consumer;
-
 	u32 last_cmd_consumer;
+
 	u32 max_tx_desc_count;
 	u32 max_rx_desc_count;
 	u32 max_jumbo_rx_desc_count;
 	u32 max_lro_rx_desc_count;
-	/* Num of instances active on cmd buffer ring */
-	u32 proc_cmd_buf_counter;
-
-	u32 num_threads, total_threads;	/*Use to keep track of xmit threads */
 
 	u32 flags;
 	u32 irq;
diff --git a/drivers/net/netxen/netxen_nic_ethtool.c b/drivers/net/netxen/netxen_nic_ethtool.c
index 7a876f4..d324ea3 100644
--- a/drivers/net/netxen/netxen_nic_ethtool.c
+++ b/drivers/net/netxen/netxen_nic_ethtool.c
@@ -70,9 +70,7 @@ static const struct netxen_nic_stats netxen_nic_gstrings_stats[] = {
 	{"uphcong", NETXEN_NIC_STAT(stats.uphcong)},
 	{"upmcong", NETXEN_NIC_STAT(stats.upmcong)},
 	{"updunno", NETXEN_NIC_STAT(stats.updunno)},
-	{"skb_freed", NETXEN_NIC_STAT(stats.skbfreed)},
 	{"tx_dropped", NETXEN_NIC_STAT(stats.txdropped)},
-	{"tx_null_skb", NETXEN_NIC_STAT(stats.txnullskb)},
 	{"csummed", NETXEN_NIC_STAT(stats.csummed)},
 	{"no_rcv", NETXEN_NIC_STAT(stats.no_rcv)},
 	{"rx_bytes", NETXEN_NIC_STAT(stats.rxbytes)},
diff --git a/drivers/net/netxen/netxen_nic_init.c b/drivers/net/netxen/netxen_nic_init.c
index 64fc18d..fe64618 100644
--- a/drivers/net/netxen/netxen_nic_init.c
+++ b/drivers/net/netxen/netxen_nic_init.c
@@ -1197,96 +1197,50 @@ u32 netxen_process_rcv_ring(struct netxen_adapter *adapter, int ctxid, int max)
 /* Process Command status ring */
 int netxen_process_cmd_ring(struct netxen_adapter *adapter)
 {
-	u32 last_consumer;
-	u32 consumer;
-	int count1 = 0;
-	int count2 = 0;
+	u32 last_consumer, consumer;
+	int count = 0, i;
 	struct netxen_cmd_buffer *buffer;
-	struct pci_dev *pdev;
+	struct pci_dev *pdev = adapter->pdev;
+	struct net_device *netdev = adapter->netdev;
 	struct netxen_skb_frag *frag;
-	u32 i;
-	int done;
+	int done = 0;
 
-	spin_lock(&adapter->tx_lock);
 	last_consumer = adapter->last_cmd_consumer;
-	DPRINTK(INFO, "procesing xmit complete\n");
-	/* we assume in this case that there is only one port and that is
-	 * port #1...changes need to be done in firmware to indicate port
-	 * number as part of the descriptor. This way we will be able to get
-	 * the netdev which is associated with that device.
-	 */
-
 	consumer = le32_to_cpu(*(adapter->cmd_consumer));
-	if (last_consumer == consumer) {	/* Ring is empty    */
-		DPRINTK(INFO, "last_consumer %d == consumer %d\n",
-			last_consumer, consumer);
-		spin_unlock(&adapter->tx_lock);
-		return 1;
-	}
-
-	adapter->proc_cmd_buf_counter++;
-	/*
-	 * Not needed - does not seem to be used anywhere.
-	 * adapter->cmd_consumer = consumer;
-	 */
-	spin_unlock(&adapter->tx_lock);
 
-	while ((last_consumer != consumer) && (count1 < MAX_STATUS_HANDLE)) {
+	while (last_consumer != consumer) {
 		buffer = &adapter->cmd_buf_arr[last_consumer];
-		pdev = adapter->pdev;
 		if (buffer->skb) {
 			frag = &buffer->frag_array[0];
 			pci_unmap_single(pdev, frag->dma, frag->length,
 					 PCI_DMA_TODEVICE);
 			frag->dma = 0ULL;
 			for (i = 1; i < buffer->frag_count; i++) {
-				DPRINTK(INFO, "getting fragment no %d\n", i);
 				frag++;	/* Get the next frag */
 				pci_unmap_page(pdev, frag->dma, frag->length,
 					       PCI_DMA_TODEVICE);
 				frag->dma = 0ULL;
 			}
 
-			adapter->stats.skbfreed++;
+			adapter->stats.xmitfinished++;
 			dev_kfree_skb_any(buffer->skb);
 			buffer->skb = NULL;
-		} else if (adapter->proc_cmd_buf_counter == 1) {
-			adapter->stats.txnullskb++;
-		}
-		if (unlikely(netif_queue_stopped(adapter->netdev)
-			     && netif_carrier_ok(adapter->netdev))
-		    && ((jiffies - adapter->netdev->trans_start) >
-			adapter->netdev->watchdog_timeo)) {
-			SCHEDULE_WORK(&adapter->tx_timeout_task);
 		}
 
 		last_consumer = get_next_index(last_consumer,
 					       adapter->max_tx_desc_count);
-		count1++;
+		if (++count >= MAX_STATUS_HANDLE)
+			break;
 	}
 
-	count2 = 0;
-	spin_lock(&adapter->tx_lock);
-	if ((--adapter->proc_cmd_buf_counter) == 0) {
+	if (count) {
 		adapter->last_cmd_consumer = last_consumer;
-		while ((adapter->last_cmd_consumer != consumer)
-		       && (count2 < MAX_STATUS_HANDLE)) {
-			buffer =
-			    &adapter->cmd_buf_arr[adapter->last_cmd_consumer];
-			count2++;
-			if (buffer->skb)
-				break;
-			else
-				adapter->last_cmd_consumer =
-				    get_next_index(adapter->last_cmd_consumer,
-						   adapter->max_tx_desc_count);
-		}
-	}
-	if (count1 || count2) {
-		if (netif_queue_stopped(adapter->netdev)
-		    && (adapter->flags & NETXEN_NETDEV_STATUS)) {
-			netif_wake_queue(adapter->netdev);
-			adapter->flags &= ~NETXEN_NETDEV_STATUS;
+		smp_mb();
+		if (netif_queue_stopped(netdev) && netif_running(netdev)) {
+			netif_tx_lock(netdev);
+			netif_wake_queue(netdev);
+			smp_mb();
+			netif_tx_unlock(netdev);
 		}
 	}
 	/*
@@ -1302,16 +1256,9 @@ int netxen_process_cmd_ring(struct netxen_adapter *adapter)
 	 * There is still a possible race condition and the host could miss an
 	 * interrupt. The card has to take care of this.
 	 */
-	if (adapter->last_cmd_consumer == consumer &&
-	    (((adapter->cmd_producer + 1) %
-	      adapter->max_tx_desc_count) == adapter->last_cmd_consumer)) {
-		consumer = le32_to_cpu(*(adapter->cmd_consumer));
-	}
-	done = (adapter->last_cmd_consumer == consumer);
+	consumer = le32_to_cpu(*(adapter->cmd_consumer));
+	done = (last_consumer == consumer);
 
-	spin_unlock(&adapter->tx_lock);
-	DPRINTK(INFO, "last consumer is %d in %s\n", last_consumer,
-		__FUNCTION__);
 	return (done);
 }
 
diff --git a/drivers/net/netxen/netxen_nic_main.c b/drivers/net/netxen/netxen_nic_main.c
index 9595520..dc4d593 100644
--- a/drivers/net/netxen/netxen_nic_main.c
+++ b/drivers/net/netxen/netxen_nic_main.c
@@ -317,7 +317,6 @@ netxen_nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 
 	adapter->ahw.pdev = pdev;
 	adapter->ahw.pci_func  = pci_func_id;
-	spin_lock_init(&adapter->tx_lock);
 
 	/* remap phys address */
 	mem_base = pci_resource_start(pdev, 0);	/* 0 is for BAR 0 */
@@ -533,7 +532,6 @@ netxen_nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	adapter->watchdog_timer.data = (unsigned long)adapter;
 	INIT_WORK(&adapter->watchdog_task, netxen_watchdog_task);
 	adapter->ahw.pdev = pdev;
-	adapter->proc_cmd_buf_counter = 0;
 	adapter->ahw.revision_id = pdev->revision;
 
 	/* make sure Window == 1 */
@@ -952,41 +950,17 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	struct netxen_skb_frag *buffrag;
 	unsigned int i;
 
-	u32 producer = 0;
+	u32 producer, consumer;
 	u32 saved_producer = 0;
 	struct cmd_desc_type0 *hwdesc;
 	int k;
 	struct netxen_cmd_buffer *pbuf = NULL;
-	static int dropped_packet = 0;
 	int frag_count;
-	u32 local_producer = 0;
-	u32 max_tx_desc_count = 0;
-	u32 last_cmd_consumer = 0;
 	int no_of_desc;
+	u32 num_txd = adapter->max_tx_desc_count;
 
-	adapter->stats.xmitcalled++;
 	frag_count = skb_shinfo(skb)->nr_frags + 1;
 
-	if (unlikely(skb->len <= 0)) {
-		dev_kfree_skb_any(skb);
-		adapter->stats.badskblen++;
-		return NETDEV_TX_OK;
-	}
-
-	if (frag_count > MAX_BUFFERS_PER_CMD) {
-		printk("%s: %s netxen_nic_xmit_frame: frag_count (%d) "
-		       "too large, can handle only %d frags\n",
-		       netxen_nic_driver_name, netdev->name,
-		       frag_count, MAX_BUFFERS_PER_CMD);
-		adapter->stats.txdropped++;
-		if ((++dropped_packet & 0xff) == 0xff)
-			printk("%s: %s droppped packets = %d\n",
-			       netxen_nic_driver_name, netdev->name,
-			       dropped_packet);
-
-		return NETDEV_TX_OK;
-	}
-
 	/* There 4 fragments per descriptor */
 	no_of_desc = (frag_count + 3) >> 2;
 	if (netdev->features & NETIF_F_TSO) {
@@ -1001,27 +975,16 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 		}
 	}
 
-	spin_lock_bh(&adapter->tx_lock);
-	if (adapter->total_threads >= MAX_XMIT_PRODUCERS) {
-		goto out_requeue;
+	producer = adapter->cmd_producer;
+	smp_mb();
+	consumer = adapter->last_cmd_consumer;
+	if ((no_of_desc+2) > find_diff_among(producer, consumer, num_txd)) {
+		netif_stop_queue(netdev);
+		smp_mb();
+		return NETDEV_TX_BUSY;
 	}
-	local_producer = adapter->cmd_producer;
-	k = adapter->cmd_producer;
-	max_tx_desc_count = adapter->max_tx_desc_count;
-	last_cmd_consumer = adapter->last_cmd_consumer;
-	if ((k + no_of_desc) >=
-	    ((last_cmd_consumer <= k) ? last_cmd_consumer + max_tx_desc_count :
-	     last_cmd_consumer)) {
-		goto out_requeue;
-	}
-	k = get_index_range(k, max_tx_desc_count, no_of_desc);
-	adapter->cmd_producer = k;
-	adapter->total_threads++;
-	adapter->num_threads++;
 
-	spin_unlock_bh(&adapter->tx_lock);
 	/* Copy the descriptors into the hardware    */
-	producer = local_producer;
 	saved_producer = producer;
 	hwdesc = &hw->cmd_desc_head[producer];
 	memset(hwdesc, 0, sizeof(struct cmd_desc_type0));
@@ -1061,8 +1024,7 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 		/* move to next desc. if there is a need */
 		if ((i & 0x3) == 0) {
 			k = 0;
-			producer = get_next_index(producer,
-						  adapter->max_tx_desc_count);
+			producer = get_next_index(producer, num_txd);
 			hwdesc = &hw->cmd_desc_head[producer];
 			memset(hwdesc, 0, sizeof(struct cmd_desc_type0));
 			pbuf = &adapter->cmd_buf_arr[producer];
@@ -1080,7 +1042,6 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 		buffrag->dma = temp_dma;
 		buffrag->length = temp_len;
 
-		DPRINTK(INFO, "for loop. i=%d k=%d\n", i, k);
 		switch (k) {
 		case 0:
 			hwdesc->buffer1_length = cpu_to_le16(temp_len);
@@ -1101,7 +1062,7 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 		}
 		frag++;
 	}
-	producer = get_next_index(producer, adapter->max_tx_desc_count);
+	producer = get_next_index(producer, num_txd);
 
 	/* might change opcode to TX_TCP_LSO */
 	netxen_tso_check(adapter, &hw->cmd_desc_head[saved_producer], skb);
@@ -1128,7 +1089,7 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 		/* copy the first 64 bytes */
 		memcpy(((void *)hwdesc) + 2,
 		       (void *)(skb->data), first_hdr_len);
-		producer = get_next_index(producer, max_tx_desc_count);
+		producer = get_next_index(producer, num_txd);
 
 		if (more_hdr) {
 			hwdesc = &hw->cmd_desc_head[producer];
@@ -1141,35 +1102,19 @@ static int netxen_nic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 							 hwdesc,
 							 (hdr_len -
 							  first_hdr_len));
-			producer = get_next_index(producer, max_tx_desc_count);
+			producer = get_next_index(producer, num_txd);
 		}
 	}
 
-	spin_lock_bh(&adapter->tx_lock);
+	adapter->cmd_producer = producer;
 	adapter->stats.txbytes += skb->len;
 
-	/* Code to update the adapter considering how many producer threads
-	   are currently working */
-	if ((--adapter->num_threads) == 0) {
-		/* This is the last thread */
-		u32 crb_producer = adapter->cmd_producer;
-		netxen_nic_update_cmd_producer(adapter, crb_producer);
-		wmb();
-		adapter->total_threads = 0;
-	}
+	netxen_nic_update_cmd_producer(adapter, adapter->cmd_producer);
 
-	adapter->stats.xmitfinished++;
+	adapter->stats.xmitcalled++;
 	netdev->trans_start = jiffies;
 
-	spin_unlock_bh(&adapter->tx_lock);
 	return NETDEV_TX_OK;
-
-out_requeue:
-	netif_stop_queue(netdev);
-	adapter->flags |= NETXEN_NETDEV_STATUS;
-
-	spin_unlock_bh(&adapter->tx_lock);
-	return NETDEV_TX_BUSY;
 }
 
 static void netxen_watchdog(unsigned long v)
@@ -1194,9 +1139,13 @@ static void netxen_tx_timeout_task(struct work_struct *work)
 	printk(KERN_ERR "%s %s: transmit timeout, resetting.\n",
 	       netxen_nic_driver_name, adapter->netdev->name);
 
-	netxen_nic_close(adapter->netdev);
-	netxen_nic_open(adapter->netdev);
+	netxen_nic_disable_int(adapter);
+	napi_disable(&adapter->napi);
+
 	adapter->netdev->trans_start = jiffies;
+
+	napi_enable(&adapter->napi);
+	netxen_nic_enable_int(adapter);
 	netif_wake_queue(adapter->netdev);
 }
 
-- 
1.5.4

--
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