[PATCH 1/5] [net-next] qlge: Clean up mac address and frame route settings.

[PATCH 1/5] [net-next] qlge: Clean up mac address and frame route settings.

[Ron Mercer]

Setting MAC addresses and routing frames to various queues will need to
be done in response to firmware events as well as during initialization.
This change encapsulates the facilities into a single call that can
later me made from other places.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |   31 ++++++++++++++++++++++---------
 1 files changed, 22 insertions(+), 9 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index fd515af..3a041b6 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -2958,6 +2958,24 @@ static int ql_route_initialize(struct ql_adapter *qdev)
 	return status;
 }
 
+static int ql_cam_route_initialize(struct ql_adapter *qdev)
+{
+	int status;
+
+	status = ql_set_mac_addr_reg(qdev, (u8 *) qdev->ndev->perm_addr,
+			     MAC_ADDR_TYPE_CAM_MAC, qdev->func * MAX_CQ);
+	if (status) {
+		QPRINTK(qdev, IFUP, ERR, "Failed to init mac address.\n");
+		return status;
+	}
+
+	status = ql_route_initialize(qdev);
+	if (status)
+		QPRINTK(qdev, IFUP, ERR, "Failed to init routing table.\n");
+
+	return status;
+}
+
 static int ql_adapter_initialize(struct ql_adapter *qdev)
 {
 	u32 value, mask;
@@ -3028,16 +3046,11 @@ static int ql_adapter_initialize(struct ql_adapter *qdev)
 		return status;
 	}
 
-	status = ql_set_mac_addr_reg(qdev, (u8 *) qdev->ndev->perm_addr,
-				     MAC_ADDR_TYPE_CAM_MAC, qdev->func);
-	if (status) {
-		QPRINTK(qdev, IFUP, ERR, "Failed to init mac address.\n");
-		return status;
-	}
-
-	status = ql_route_initialize(qdev);
+	/* Set up the MAC address and frame routing filter. */
+	status = ql_cam_route_initialize(qdev);
 	if (status) {
-		QPRINTK(qdev, IFUP, ERR, "Failed to init routing table.\n");
+		QPRINTK(qdev, IFUP, ERR,
+				"Failed to init CAM/Routing tables.\n");
 		return status;
 	}
 
-- 
1.6.0.2

[PATCH 2/5] [net-next] qlge: Increase MAC addr hw sem granularity.

[Ron Mercer]

Instead of taking/giving the semaphore repeatedly when iterating over
several adderesses, we have the caller hold it until all are done.  This
reduces PCI bus chatter and possible waits.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_dbg.c  |    9 +++++++
 drivers/net/qlge/qlge_main.c |   51 ++++++++++++++++++++++++++++--------------
 2 files changed, 43 insertions(+), 17 deletions(-)

diff --git a/drivers/net/qlge/qlge_dbg.c b/drivers/net/qlge/qlge_dbg.c
index 379b895..40a70c3 100644
--- a/drivers/net/qlge/qlge_dbg.c
+++ b/drivers/net/qlge/qlge_dbg.c
@@ -83,6 +83,10 @@ static void ql_dump_cam_entries(struct ql_adapter *qdev)
 {
 	int i;
 	u32 value[3];
+
+	i = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+	if (i)
+		return;
 	for (i = 0; i < 4; i++) {
 		if (ql_get_mac_addr_reg(qdev, MAC_ADDR_TYPE_CAM_MAC, i, value)) {
 			printk(KERN_ERR PFX
@@ -111,12 +115,16 @@ static void ql_dump_cam_entries(struct ql_adapter *qdev)
 				       qdev->ndev->name, i, value[1], value[0]);
 		}
 	}
+	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 }
 
 void ql_dump_routing_entries(struct ql_adapter *qdev)
 {
 	int i;
 	u32 value;
+	i = ql_sem_spinlock(qdev, SEM_RT_IDX_MASK);
+	if (i)
+		return;
 	for (i = 0; i < 16; i++) {
 		value = 0;
 		if (ql_get_routing_reg(qdev, i, &value)) {
@@ -131,6 +139,7 @@ void ql_dump_routing_entries(struct ql_adapter *qdev)
 				       qdev->ndev->name, i, value);
 		}
 	}
+	ql_sem_unlock(qdev, SEM_RT_IDX_MASK);
 }
 
 void ql_dump_regs(struct ql_adapter *qdev)
diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 3a041b6..d316e50 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -247,9 +247,6 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 	u32 offset = 0;
 	int status;
 
-	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
-	if (status)
-		return status;
 	switch (type) {
 	case MAC_ADDR_TYPE_MULTI_MAC:
 	case MAC_ADDR_TYPE_CAM_MAC:
@@ -308,7 +305,6 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 		status = -EPERM;
 	}
 exit:
-	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 	return status;
 }
 
@@ -321,9 +317,6 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 	u32 offset = 0;
 	int status = 0;
 
-	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
-	if (status)
-		return status;
 	switch (type) {
 	case MAC_ADDR_TYPE_MULTI_MAC:
 	case MAC_ADDR_TYPE_CAM_MAC:
@@ -415,7 +408,6 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 		status = -EPERM;
 	}
 exit:
-	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 	return status;
 }
 
@@ -1690,19 +1682,29 @@ static void ql_vlan_rx_add_vid(struct net_device *ndev, u16 vid)
 {
 	struct ql_adapter *qdev = netdev_priv(ndev);
 	u32 enable_bit = MAC_ADDR_E;
+	int status;
 
+	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+	if (status)
+		return;
 	spin_lock(&qdev->hw_lock);
 	if (ql_set_mac_addr_reg
 	    (qdev, (u8 *) &enable_bit, MAC_ADDR_TYPE_VLAN, vid)) {
 		QPRINTK(qdev, IFUP, ERR, "Failed to init vlan address.\n");
 	}
 	spin_unlock(&qdev->hw_lock);
+	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 }
 
 static void ql_vlan_rx_kill_vid(struct net_device *ndev, u16 vid)
 {
 	struct ql_adapter *qdev = netdev_priv(ndev);
 	u32 enable_bit = 0;
+	int status;
+
+	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+	if (status)
+		return;
 
 	spin_lock(&qdev->hw_lock);
 	if (ql_set_mac_addr_reg
@@ -1710,6 +1712,7 @@ static void ql_vlan_rx_kill_vid(struct net_device *ndev, u16 vid)
 		QPRINTK(qdev, IFUP, ERR, "Failed to clear vlan address.\n");
 	}
 	spin_unlock(&qdev->hw_lock);
+	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 
 }
 
@@ -2962,8 +2965,12 @@ static int ql_cam_route_initialize(struct ql_adapter *qdev)
 {
 	int status;
 
+	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+	if (status)
+		return status;
 	status = ql_set_mac_addr_reg(qdev, (u8 *) qdev->ndev->perm_addr,
 			     MAC_ADDR_TYPE_CAM_MAC, qdev->func * MAX_CQ);
+	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 	if (status) {
 		QPRINTK(qdev, IFUP, ERR, "Failed to init mac address.\n");
 		return status;
@@ -3428,8 +3435,11 @@ static void qlge_set_multicast_list(struct net_device *ndev)
 {
 	struct ql_adapter *qdev = (struct ql_adapter *)netdev_priv(ndev);
 	struct dev_mc_list *mc_ptr;
-	int i;
+	int i, status;
 
+	status = ql_sem_spinlock(qdev, SEM_RT_IDX_MASK);
+	if (status)
+		return;
 	spin_lock(&qdev->hw_lock);
 	/*
 	 * Set or clear promiscuous mode if a
@@ -3485,14 +3495,19 @@ static void qlge_set_multicast_list(struct net_device *ndev)
 	}
 
 	if (ndev->mc_count) {
+		status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+		if (status)
+			goto exit;
 		for (i = 0, mc_ptr = ndev->mc_list; mc_ptr;
 		     i++, mc_ptr = mc_ptr->next)
 			if (ql_set_mac_addr_reg(qdev, (u8 *) mc_ptr->dmi_addr,
 						MAC_ADDR_TYPE_MULTI_MAC, i)) {
 				QPRINTK(qdev, HW, ERR,
 					"Failed to loadmulticast address.\n");
+				ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 				goto exit;
 			}
+		ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
 		if (ql_set_routing_reg
 		    (qdev, RT_IDX_MCAST_MATCH_SLOT, RT_IDX_MCAST_MATCH, 1)) {
 			QPRINTK(qdev, HW, ERR,
@@ -3509,7 +3524,7 @@ static int qlge_set_mac_address(struct net_device *ndev, void *p)
 {
 	struct ql_adapter *qdev = (struct ql_adapter *)netdev_priv(ndev);
 	struct sockaddr *addr = p;
-	int ret = 0;
+	int status;
 
 	if (netif_running(ndev))
 		return -EBUSY;
@@ -3518,15 +3533,17 @@ static int qlge_set_mac_address(struct net_device *ndev, void *p)
 		return -EADDRNOTAVAIL;
 	memcpy(ndev->dev_addr, addr->sa_data, ndev->addr_len);
 
+	status = ql_sem_spinlock(qdev, SEM_MAC_ADDR_MASK);
+	if (status)
+		return status;
 	spin_lock(&qdev->hw_lock);
-	if (ql_set_mac_addr_reg(qdev, (u8 *) ndev->dev_addr,
-			MAC_ADDR_TYPE_CAM_MAC, qdev->func)) {/* Unicast */
-		QPRINTK(qdev, HW, ERR, "Failed to load MAC address.\n");
-		ret = -1;
-	}
+	status = ql_set_mac_addr_reg(qdev, (u8 *) ndev->dev_addr,
+			MAC_ADDR_TYPE_CAM_MAC, qdev->func * MAX_CQ);
 	spin_unlock(&qdev->hw_lock);
-
-	return ret;
+	if (status)
+		QPRINTK(qdev, HW, ERR, "Failed to load MAC address.\n");
+	ql_sem_unlock(qdev, SEM_MAC_ADDR_MASK);
+	return status;
 }
 
 static void qlge_tx_timeout(struct net_device *ndev)
-- 
1.6.0.2

[PATCH 3/5] [net-next] qlge: Change frame route hw semaphore granularity.

[Ron Mercer]

Instead of taking/giving the hw semaphore repeatedly when iterating over
several frame to queue route settings, we have the caller hold it until
all are done.
This reduces PCI bus chatter and possible waits.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |   31 +++++++++++++------------------
 1 files changed, 13 insertions(+), 18 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index d316e50..16c056a 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -418,10 +418,6 @@ int ql_get_routing_reg(struct ql_adapter *qdev, u32 index, u32 *value)
 {
 	int status = 0;
 
-	status = ql_sem_spinlock(qdev, SEM_RT_IDX_MASK);
-	if (status)
-		goto exit;
-
 	status = ql_wait_reg_rdy(qdev, RT_IDX, RT_IDX_MW, 0);
 	if (status)
 		goto exit;
@@ -433,7 +429,6 @@ int ql_get_routing_reg(struct ql_adapter *qdev, u32 index, u32 *value)
 		goto exit;
 	*value = ql_read32(qdev, RT_DATA);
 exit:
-	ql_sem_unlock(qdev, SEM_RT_IDX_MASK);
 	return status;
 }
 
@@ -445,13 +440,9 @@ exit:
 static int ql_set_routing_reg(struct ql_adapter *qdev, u32 index, u32 mask,
 			      int enable)
 {
-	int status;
+	int status = -EINVAL; /* Return error if no mask match. */
 	u32 value = 0;
 
-	status = ql_sem_spinlock(qdev, SEM_RT_IDX_MASK);
-	if (status)
-		return status;
-
 	QPRINTK(qdev, IFUP, DEBUG,
 		"%s %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s mask %s the routing reg.\n",
 		(enable ? "Adding" : "Removing"),
@@ -547,7 +538,6 @@ static int ql_set_routing_reg(struct ql_adapter *qdev, u32 index, u32 mask,
 		ql_write32(qdev, RT_DATA, enable ? mask : 0);
 	}
 exit:
-	ql_sem_unlock(qdev, SEM_RT_IDX_MASK);
 	return status;
 }
 
@@ -2916,13 +2906,17 @@ static int ql_route_initialize(struct ql_adapter *qdev)
 	int status = 0;
 	int i;
 
+	status = ql_sem_spinlock(qdev, SEM_RT_IDX_MASK);
+	if (status)
+		return status;
+
 	/* Clear all the entries in the routing table. */
 	for (i = 0; i < 16; i++) {
 		status = ql_set_routing_reg(qdev, i, 0, 0);
 		if (status) {
 			QPRINTK(qdev, IFUP, ERR,
 				"Failed to init routing register for CAM packets.\n");
-			return status;
+			goto exit;
 		}
 	}
 
@@ -2930,13 +2924,13 @@ static int ql_route_initialize(struct ql_adapter *qdev)
 	if (status) {
 		QPRINTK(qdev, IFUP, ERR,
 			"Failed to init routing register for error packets.\n");
-		return status;
+		goto exit;
 	}
 	status = ql_set_routing_reg(qdev, RT_IDX_BCAST_SLOT, RT_IDX_BCAST, 1);
 	if (status) {
 		QPRINTK(qdev, IFUP, ERR,
 			"Failed to init routing register for broadcast packets.\n");
-		return status;
+		goto exit;
 	}
 	/* If we have more than one inbound queue, then turn on RSS in the
 	 * routing block.
@@ -2947,17 +2941,17 @@ static int ql_route_initialize(struct ql_adapter *qdev)
 		if (status) {
 			QPRINTK(qdev, IFUP, ERR,
 				"Failed to init routing register for MATCH RSS packets.\n");
-			return status;
+			goto exit;
 		}
 	}
 
 	status = ql_set_routing_reg(qdev, RT_IDX_CAM_HIT_SLOT,
 				    RT_IDX_CAM_HIT, 1);
-	if (status) {
+	if (status)
 		QPRINTK(qdev, IFUP, ERR,
 			"Failed to init routing register for CAM packets.\n");
-		return status;
-	}
+exit:
+	ql_sem_unlock(qdev, SEM_RT_IDX_MASK);
 	return status;
 }
 
@@ -3518,6 +3512,7 @@ static void qlge_set_multicast_list(struct net_device *ndev)
 	}
 exit:
 	spin_unlock(&qdev->hw_lock);
+	ql_sem_unlock(qdev, SEM_RT_IDX_MASK);
 }
 
 static int qlge_set_mac_address(struct net_device *ndev, void *p)
-- 
1.6.0.2

[PATCH 4/5] [net-next] qlge: Optimize rx buffer refill process.

[Ron Mercer]

RX Buffers are refilled in chunks of 16 at a time before notifying the
hardware with a register write.  This can cause several writes to take
place in a given napi poll call.  This change causes the write to take place
only once at the end of the call.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |   17 ++++++++++++-----
 1 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 16c056a..1719845 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -863,7 +863,8 @@ static void ql_write_cq_idx(struct rx_ring *rx_ring)
 /* Process (refill) a large buffer queue. */
 static void ql_update_lbq(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 {
-	int clean_idx = rx_ring->lbq_clean_idx;
+	u32 clean_idx = rx_ring->lbq_clean_idx;
+	u32 start_idx = clean_idx;
 	struct bq_desc *lbq_desc;
 	u64 map;
 	int i;
@@ -910,19 +911,23 @@ static void ql_update_lbq(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		rx_ring->lbq_prod_idx += 16;
 		if (rx_ring->lbq_prod_idx == rx_ring->lbq_len)
 			rx_ring->lbq_prod_idx = 0;
+		rx_ring->lbq_free_cnt -= 16;
+	}
+
+	if (start_idx != clean_idx) {
 		QPRINTK(qdev, RX_STATUS, DEBUG,
 			"lbq: updating prod idx = %d.\n",
 			rx_ring->lbq_prod_idx);
 		ql_write_db_reg(rx_ring->lbq_prod_idx,
 				rx_ring->lbq_prod_idx_db_reg);
-		rx_ring->lbq_free_cnt -= 16;
 	}
 }
 
 /* Process (refill) a small buffer queue. */
 static void ql_update_sbq(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 {
-	int clean_idx = rx_ring->sbq_clean_idx;
+	u32 clean_idx = rx_ring->sbq_clean_idx;
+	u32 start_idx = clean_idx;
 	struct bq_desc *sbq_desc;
 	u64 map;
 	int i;
@@ -972,13 +977,15 @@ static void ql_update_sbq(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		rx_ring->sbq_prod_idx += 16;
 		if (rx_ring->sbq_prod_idx == rx_ring->sbq_len)
 			rx_ring->sbq_prod_idx = 0;
+		rx_ring->sbq_free_cnt -= 16;
+	}
+
+	if (start_idx != clean_idx) {
 		QPRINTK(qdev, RX_STATUS, DEBUG,
 			"sbq: updating prod idx = %d.\n",
 			rx_ring->sbq_prod_idx);
 		ql_write_db_reg(rx_ring->sbq_prod_idx,
 				rx_ring->sbq_prod_idx_db_reg);
-
-		rx_ring->sbq_free_cnt -= 16;
 	}
 }
 
-- 
1.6.0.2

[PATCH 5/5] [net-next] qlge: Use one path to (re)fill rx buffers.

[Ron Mercer]

Currently there are two paths for filling rx buffer queues. One is
used during initialization and the other during runtime.  This patch
removes ql_alloc_sbq_buffers() and ql_alloc_lbq_buffers() and replaces
them with a call to the runtime functions ql_update_lbq() and
ql_update_sbq().

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |  170 +++++++++++++++---------------------------
 1 files changed, 59 insertions(+), 111 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 1719845..655f3c4 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -2105,47 +2105,6 @@ static void ql_free_lbq_buffers(struct ql_adapter *qdev, struct rx_ring *rx_ring
 	}
 }
 
-/*
- * Allocate and map a page for each element of the lbq.
- */
-static int ql_alloc_lbq_buffers(struct ql_adapter *qdev,
-				struct rx_ring *rx_ring)
-{
-	int i;
-	struct bq_desc *lbq_desc;
-	u64 map;
-	__le64 *bq = rx_ring->lbq_base;
-
-	for (i = 0; i < rx_ring->lbq_len; i++) {
-		lbq_desc = &rx_ring->lbq[i];
-		memset(lbq_desc, 0, sizeof(lbq_desc));
-		lbq_desc->addr = bq;
-		lbq_desc->index = i;
-		lbq_desc->p.lbq_page = alloc_page(GFP_ATOMIC);
-		if (unlikely(!lbq_desc->p.lbq_page)) {
-			QPRINTK(qdev, IFUP, ERR, "failed alloc_page().\n");
-			goto mem_error;
-		} else {
-			map = pci_map_page(qdev->pdev,
-					   lbq_desc->p.lbq_page,
-					   0, PAGE_SIZE, PCI_DMA_FROMDEVICE);
-			if (pci_dma_mapping_error(qdev->pdev, map)) {
-				QPRINTK(qdev, IFUP, ERR,
-					"PCI mapping failed.\n");
-				goto mem_error;
-			}
-			pci_unmap_addr_set(lbq_desc, mapaddr, map);
-			pci_unmap_len_set(lbq_desc, maplen, PAGE_SIZE);
-			*lbq_desc->addr = cpu_to_le64(map);
-		}
-		bq++;
-	}
-	return 0;
-mem_error:
-	ql_free_lbq_buffers(qdev, rx_ring);
-	return -ENOMEM;
-}
-
 static void ql_free_sbq_buffers(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 {
 	int i;
@@ -2168,63 +2127,72 @@ static void ql_free_sbq_buffers(struct ql_adapter *qdev, struct rx_ring *rx_ring
 	}
 }
 
-/* Allocate and map an skb for each element of the sbq. */
-static int ql_alloc_sbq_buffers(struct ql_adapter *qdev,
+/* Free all large and small rx buffers associated
+ * with the completion queues for this device.
+ */
+static void ql_free_rx_buffers(struct ql_adapter *qdev)
+{
+	int i;
+	struct rx_ring *rx_ring;
+
+	for (i = 0; i < qdev->rx_ring_count; i++) {
+		rx_ring = &qdev->rx_ring[i];
+		if (rx_ring->lbq)
+			ql_free_lbq_buffers(qdev, rx_ring);
+		if (rx_ring->sbq)
+			ql_free_sbq_buffers(qdev, rx_ring);
+	}
+}
+
+static void ql_alloc_rx_buffers(struct ql_adapter *qdev)
+{
+	struct rx_ring *rx_ring;
+	int i;
+
+	for (i = 0; i < qdev->rx_ring_count; i++) {
+		rx_ring = &qdev->rx_ring[i];
+		if (rx_ring->type != TX_Q)
+			ql_update_buffer_queues(qdev, rx_ring);
+	}
+}
+
+static void ql_init_lbq_ring(struct ql_adapter *qdev,
+				struct rx_ring *rx_ring)
+{
+	int i;
+	struct bq_desc *lbq_desc;
+	__le64 *bq = rx_ring->lbq_base;
+
+	memset(rx_ring->lbq, 0, rx_ring->lbq_len * sizeof(struct bq_desc));
+	for (i = 0; i < rx_ring->lbq_len; i++) {
+		lbq_desc = &rx_ring->lbq[i];
+		memset(lbq_desc, 0, sizeof(*lbq_desc));
+		lbq_desc->index = i;
+		lbq_desc->addr = bq;
+		bq++;
+	}
+}
+
+static void ql_init_sbq_ring(struct ql_adapter *qdev,
 				struct rx_ring *rx_ring)
 {
 	int i;
 	struct bq_desc *sbq_desc;
-	struct sk_buff *skb;
-	u64 map;
 	__le64 *bq = rx_ring->sbq_base;
 
+	memset(rx_ring->sbq, 0, rx_ring->sbq_len * sizeof(struct bq_desc));
 	for (i = 0; i < rx_ring->sbq_len; i++) {
 		sbq_desc = &rx_ring->sbq[i];
-		memset(sbq_desc, 0, sizeof(sbq_desc));
+		memset(sbq_desc, 0, sizeof(*sbq_desc));
 		sbq_desc->index = i;
 		sbq_desc->addr = bq;
-		skb = netdev_alloc_skb(qdev->ndev, rx_ring->sbq_buf_size);
-		if (unlikely(!skb)) {
-			/* Better luck next round */
-			QPRINTK(qdev, IFUP, ERR,
-				"small buff alloc failed for %d bytes at index %d.\n",
-				rx_ring->sbq_buf_size, i);
-			goto mem_err;
-		}
-		skb_reserve(skb, QLGE_SB_PAD);
-		sbq_desc->p.skb = skb;
-		/*
-		 * Map only half the buffer. Because the
-		 * other half may get some data copied to it
-		 * when the completion arrives.
-		 */
-		map = pci_map_single(qdev->pdev,
-				     skb->data,
-				     rx_ring->sbq_buf_size / 2,
-				     PCI_DMA_FROMDEVICE);
-		if (pci_dma_mapping_error(qdev->pdev, map)) {
-			QPRINTK(qdev, IFUP, ERR, "PCI mapping failed.\n");
-			goto mem_err;
-		}
-		pci_unmap_addr_set(sbq_desc, mapaddr, map);
-		pci_unmap_len_set(sbq_desc, maplen, rx_ring->sbq_buf_size / 2);
-		*sbq_desc->addr = cpu_to_le64(map);
 		bq++;
 	}
-	return 0;
-mem_err:
-	ql_free_sbq_buffers(qdev, rx_ring);
-	return -ENOMEM;
 }
 
 static void ql_free_rx_resources(struct ql_adapter *qdev,
 				 struct rx_ring *rx_ring)
 {
-	if (rx_ring->sbq_len)
-		ql_free_sbq_buffers(qdev, rx_ring);
-	if (rx_ring->lbq_len)
-		ql_free_lbq_buffers(qdev, rx_ring);
-
 	/* Free the small buffer queue. */
 	if (rx_ring->sbq_base) {
 		pci_free_consistent(qdev->pdev,
@@ -2302,11 +2270,7 @@ static int ql_alloc_rx_resources(struct ql_adapter *qdev,
 			goto err_mem;
 		}
 
-		if (ql_alloc_sbq_buffers(qdev, rx_ring)) {
-			QPRINTK(qdev, IFUP, ERR,
-				"Small buffer allocation failed.\n");
-			goto err_mem;
-		}
+		ql_init_sbq_ring(qdev, rx_ring);
 	}
 
 	if (rx_ring->lbq_len) {
@@ -2334,14 +2298,7 @@ static int ql_alloc_rx_resources(struct ql_adapter *qdev,
 			goto err_mem;
 		}
 
-		/*
-		 * Allocate the buffers.
-		 */
-		if (ql_alloc_lbq_buffers(qdev, rx_ring)) {
-			QPRINTK(qdev, IFUP, ERR,
-				"Large buffer allocation failed.\n");
-			goto err_mem;
-		}
+		ql_init_lbq_ring(qdev, rx_ring);
 	}
 
 	return 0;
@@ -2489,10 +2446,10 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		bq_len = (rx_ring->lbq_len == 65536) ? 0 :
 			(u16) rx_ring->lbq_len;
 		cqicb->lbq_len = cpu_to_le16(bq_len);
-		rx_ring->lbq_prod_idx = rx_ring->lbq_len - 16;
+		rx_ring->lbq_prod_idx = 0;
 		rx_ring->lbq_curr_idx = 0;
-		rx_ring->lbq_clean_idx = rx_ring->lbq_prod_idx;
-		rx_ring->lbq_free_cnt = 16;
+		rx_ring->lbq_clean_idx = 0;
+		rx_ring->lbq_free_cnt = rx_ring->lbq_len;
 	}
 	if (rx_ring->sbq_len) {
 		cqicb->flags |= FLAGS_LS;	/* Load sbq values */
@@ -2504,10 +2461,10 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		bq_len = (rx_ring->sbq_len == 65536) ? 0 :
 			(u16) rx_ring->sbq_len;
 		cqicb->sbq_len = cpu_to_le16(bq_len);
-		rx_ring->sbq_prod_idx = rx_ring->sbq_len - 16;
+		rx_ring->sbq_prod_idx = 0;
 		rx_ring->sbq_curr_idx = 0;
-		rx_ring->sbq_clean_idx = rx_ring->sbq_prod_idx;
-		rx_ring->sbq_free_cnt = 16;
+		rx_ring->sbq_clean_idx = 0;
+		rx_ring->sbq_free_cnt = rx_ring->sbq_len;
 	}
 	switch (rx_ring->type) {
 	case TX_Q:
@@ -2560,17 +2517,6 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		QPRINTK(qdev, IFUP, ERR, "Failed to load CQICB.\n");
 		return err;
 	}
-	QPRINTK(qdev, IFUP, INFO, "Successfully loaded CQICB.\n");
-	/*
-	 * Advance the producer index for the buffer queues.
-	 */
-	wmb();
-	if (rx_ring->lbq_len)
-		ql_write_db_reg(rx_ring->lbq_prod_idx,
-				rx_ring->lbq_prod_idx_db_reg);
-	if (rx_ring->sbq_len)
-		ql_write_db_reg(rx_ring->sbq_prod_idx,
-				rx_ring->sbq_prod_idx_db_reg);
 	return err;
 }
 
@@ -3171,6 +3117,7 @@ static int ql_adapter_down(struct ql_adapter *qdev)
 
 	ql_tx_ring_clean(qdev);
 
+	ql_free_rx_buffers(qdev);
 	spin_lock(&qdev->hw_lock);
 	status = ql_adapter_reset(qdev);
 	if (status)
@@ -3193,6 +3140,7 @@ static int ql_adapter_up(struct ql_adapter *qdev)
 	}
 	spin_unlock(&qdev->hw_lock);
 	set_bit(QL_ADAPTER_UP, &qdev->flags);
+	ql_alloc_rx_buffers(qdev);
 	ql_enable_interrupts(qdev);
 	ql_enable_all_completion_interrupts(qdev);
 	if ((ql_read32(qdev, STS) & qdev->port_init)) {
-- 
1.6.0.2

Re: [PATCH 1/5] [net-next] qlge: Clean up mac address and frame route settings.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Mon, 23 Feb 2009 12:42:13 -0800

> Setting MAC addresses and routing frames to various queues will need to
> be done in response to firmware events as well as during initialization.
> This change encapsulates the facilities into a single call that can
> later me made from other places.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

Re: [PATCH 2/5] [net-next] qlge: Increase MAC addr hw sem granularity.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Mon, 23 Feb 2009 12:42:14 -0800

> Instead of taking/giving the semaphore repeatedly when iterating over
> several adderesses, we have the caller hold it until all are done.  This
> reduces PCI bus chatter and possible waits.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

Re: [PATCH 3/5] [net-next] qlge: Change frame route hw semaphore granularity.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Mon, 23 Feb 2009 12:42:15 -0800

> Instead of taking/giving the hw semaphore repeatedly when iterating over
> several frame to queue route settings, we have the caller hold it until
> all are done.
> This reduces PCI bus chatter and possible waits.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

Re: [PATCH 4/5] [net-next] qlge: Optimize rx buffer refill process.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Mon, 23 Feb 2009 12:42:16 -0800

> RX Buffers are refilled in chunks of 16 at a time before notifying the
> hardware with a register write.  This can cause several writes to take
> place in a given napi poll call.  This change causes the write to take place
> only once at the end of the call.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

Re: [PATCH 5/5] [net-next] qlge: Use one path to (re)fill rx buffers.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Mon, 23 Feb 2009 12:42:17 -0800

> Currently there are two paths for filling rx buffer queues. One is
> used during initialization and the other during runtime.  This patch
> removes ql_alloc_sbq_buffers() and ql_alloc_lbq_buffers() and replaces
> them with a call to the runtime functions ql_update_lbq() and
> ql_update_sbq().
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.