[PATCH 1/6] ibmveth: Enable TCP checksum offload

[PATCH 1/6] ibmveth: Enable TCP checksum offload

[Brian King]

This patchset enables TCP checksum offload support for IPV4
on ibmveth. This completely eliminates the generation and checking of
the checksum for packets that are completely virtual and never
touch a physical network. A simple TCP_STREAM netperf run on
a virtual network with maximum mtu set yielded a ~30% increase
in throughput. This feature is enabled by default on systems that
support it, but can be disabled with a module option.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |   53 ++++++++++++++++++++++++++++++++
 linux-2.6-bjking1/drivers/net/ibmveth.h |   41 +++++++++++++++++++++++-
 2 files changed, 92 insertions(+), 2 deletions(-)

diff -puN drivers/net/ibmveth.c~ibmveth_csum_offload drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_csum_offload	2007-08-01 14:55:11.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-01 14:55:11.000000000 -0500
@@ -47,6 +47,8 @@
 #include <linux/mm.h>
 #include <linux/ethtool.h>
 #include <linux/proc_fs.h>
+#include <linux/in.h>
+#include <linux/ip.h>
 #include <asm/semaphore.h>
 #include <asm/hvcall.h>
 #include <asm/atomic.h>
@@ -131,6 +133,11 @@ static inline int ibmveth_rxq_frame_leng
 	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 }
 
+static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
+{
+	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].csum_good);
+}
+
 /* setup the initial settings for a buffer pool */
 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 {
@@ -684,6 +691,24 @@ static int ibmveth_start_xmit(struct sk_
 					desc[0].fields.length, DMA_TO_DEVICE);
 	desc[0].fields.valid   = 1;
 
+	if (skb->ip_summed == CHECKSUM_PARTIAL &&
+	    ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
+		ibmveth_error_printk("tx: failed to checksum packet\n");
+		tx_dropped++;
+		goto out;
+	}
+
+	if (skb->ip_summed == CHECKSUM_PARTIAL) {
+		unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
+
+		desc[0].fields.no_csum = 1;
+		desc[0].fields.csum_good = 1;
+
+		/* Need to zero out the checksum */
+		buf[0] = 0;
+		buf[1] = 0;
+	}
+
 	if(dma_mapping_error(desc[0].fields.address)) {
 		ibmveth_error_printk("tx: unable to map initial fragment\n");
 		tx_map_failed++;
@@ -702,6 +727,10 @@ static int ibmveth_start_xmit(struct sk_
 				frag->size, DMA_TO_DEVICE);
 		desc[curfrag+1].fields.length = frag->size;
 		desc[curfrag+1].fields.valid  = 1;
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			desc[curfrag+1].fields.no_csum = 1;
+			desc[curfrag+1].fields.csum_good = 1;
+		}
 
 		if(dma_mapping_error(desc[curfrag+1].fields.address)) {
 			ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
@@ -792,7 +821,11 @@ static int ibmveth_poll(struct net_devic
 			} else {
 				int length = ibmveth_rxq_frame_length(adapter);
 				int offset = ibmveth_rxq_frame_offset(adapter);
+				int csum_good = ibmveth_rxq_csum_good(adapter);
+
 				skb = ibmveth_rxq_get_buffer(adapter);
+				if (csum_good)
+					skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 				ibmveth_rxq_harvest_buffer(adapter);
 
@@ -962,8 +995,10 @@ static void ibmveth_poll_controller(stru
 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 {
 	int rc, i;
+	long ret;
 	struct net_device *netdev;
 	struct ibmveth_adapter *adapter = NULL;
+	union ibmveth_illan_attributes set_attr, ret_attr;
 
 	unsigned char *mac_addr_p;
 	unsigned int *mcastFilterSize_p;
@@ -1058,6 +1093,24 @@ static int __devinit ibmveth_probe(struc
 
 	ibmveth_debug_printk("registering netdev...\n");
 
+	ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr.desc);
+
+	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
+	    !ret_attr.fields.trunk_priority &&
+	    ret_attr.fields.csum_offload_padded_pkt_support) {
+		set_attr.desc = 0;
+		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+
+		ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
+					 &ret_attr.desc);
+
+		if (ret == H_SUCCESS)
+			netdev->features |= NETIF_F_IP_CSUM;
+		else
+			ret = h_illan_attributes(dev->unit_address, set_attr.desc,
+						 0, &ret_attr.desc);
+	}
+
 	rc = register_netdev(netdev);
 
 	if(rc) {
diff -puN drivers/net/ibmveth.h~ibmveth_csum_offload drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_csum_offload	2007-08-01 14:55:11.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-01 14:55:11.000000000 -0500
@@ -67,6 +67,21 @@ static inline long h_send_logical_lan(un
 	return rc;
 }
 
+static inline long h_illan_attributes(unsigned long unit_address,
+				      unsigned long reset_mask, unsigned long set_mask,
+				      unsigned long *ret_attributes)
+{
+	long rc;
+	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
+
+	rc = plpar_hcall(H_ILLAN_ATTRIBUTES, retbuf, unit_address,
+			 reset_mask, set_mask);
+
+	*ret_attributes = retbuf[0];
+
+	return rc;
+}
+
 #define h_multicast_ctrl(ua, cmd, mac) \
   plpar_hcall_norets(H_MULTICAST_CTRL, ua, cmd, mac)
 
@@ -141,7 +156,9 @@ struct ibmveth_adapter {
 struct ibmveth_buf_desc_fields {
     u32 valid : 1;
     u32 toggle : 1;
-    u32 reserved : 6;
+    u32 reserved : 4;
+    u32 no_csum : 1;
+    u32 csum_good : 1;
     u32 length : 24;
     u32 address;
 };
@@ -151,10 +168,30 @@ union ibmveth_buf_desc {
     struct ibmveth_buf_desc_fields fields;
 };
 
+struct ibmveth_illan_attributes_fields {
+	u32 reserved;
+	u32 reserved2 : 18;
+	u32 csum_offload_padded_pkt_support : 1;
+	u32 reserved3 : 1;
+	u32 trunk_priority : 4;
+	u32 reserved4 : 5;
+	u32 tcp_csum_offload_ipv6 : 1;
+	u32 tcp_csum_offload_ipv4 : 1;
+	u32 active_trunk : 1;
+};
+
+union ibmveth_illan_attributes {
+	u64 desc;
+	struct ibmveth_illan_attributes_fields fields;
+};
+
 struct ibmveth_rx_q_entry {
     u16 toggle : 1;
     u16 valid : 1;
-    u16 reserved : 14;
+    u16 reserved : 4;
+    u16 no_csum : 1;
+    u16 csum_good : 1;
+    u16 reserved2 : 8;
     u16 offset;
     u32 length;
     u64 correlator;
_

Re: [PATCH 1/6] ibmveth: Enable TCP checksum offload

[Jeff Garzik]

Brian King wrote:
> This patchset enables TCP checksum offload support for IPV4
> on ibmveth. This completely eliminates the generation and checking of
> the checksum for packets that are completely virtual and never
> touch a physical network. A simple TCP_STREAM netperf run on
> a virtual network with maximum mtu set yielded a ~30% increase
> in throughput. This feature is enabled by default on systems that
> support it, but can be disabled with a module option.
> 
> Signed-off-by: Brian King <brking@linux.vnet.ibm.com>

ACK, but does not apply to current netdev-2.6.git#upstream-fixes. 
Request resend after the ibmveth fixes hit mainline (24 hours or so 
after push, I suppose)

[PATCH 1/6] ibmveth: Enable TCP checksum offload

[Brian King]

This patchset enables TCP checksum offload support for IPV4
on ibmveth. This completely eliminates the generation and checking of
the checksum for packets that are completely virtual and never
touch a physical network. A simple TCP_STREAM netperf run on
a virtual network with maximum mtu set yielded a ~30% increase
in throughput. This feature is enabled by default on systems that
support it, but can be disabled with a module option.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |   53 ++++++++++++++++++++++++++++++++
 linux-2.6-bjking1/drivers/net/ibmveth.h |   41 +++++++++++++++++++++++-
 2 files changed, 92 insertions(+), 2 deletions(-)

diff -puN drivers/net/ibmveth.c~ibmveth_csum_offload drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_csum_offload	2007-08-09 13:08:21.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-09 13:08:21.000000000 -0500
@@ -47,6 +47,8 @@
 #include <linux/mm.h>
 #include <linux/ethtool.h>
 #include <linux/proc_fs.h>
+#include <linux/in.h>
+#include <linux/ip.h>
 #include <asm/semaphore.h>
 #include <asm/hvcall.h>
 #include <asm/atomic.h>
@@ -131,6 +133,11 @@ static inline int ibmveth_rxq_frame_leng
 	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 }
 
+static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
+{
+	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].csum_good);
+}
+
 /* setup the initial settings for a buffer pool */
 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 {
@@ -684,6 +691,24 @@ static int ibmveth_start_xmit(struct sk_
 					desc[0].fields.length, DMA_TO_DEVICE);
 	desc[0].fields.valid   = 1;
 
+	if (skb->ip_summed == CHECKSUM_PARTIAL &&
+	    ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
+		ibmveth_error_printk("tx: failed to checksum packet\n");
+		tx_dropped++;
+		goto out;
+	}
+
+	if (skb->ip_summed == CHECKSUM_PARTIAL) {
+		unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
+
+		desc[0].fields.no_csum = 1;
+		desc[0].fields.csum_good = 1;
+
+		/* Need to zero out the checksum */
+		buf[0] = 0;
+		buf[1] = 0;
+	}
+
 	if(dma_mapping_error(desc[0].fields.address)) {
 		ibmveth_error_printk("tx: unable to map initial fragment\n");
 		tx_map_failed++;
@@ -702,6 +727,10 @@ static int ibmveth_start_xmit(struct sk_
 				frag->size, DMA_TO_DEVICE);
 		desc[curfrag+1].fields.length = frag->size;
 		desc[curfrag+1].fields.valid  = 1;
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			desc[curfrag+1].fields.no_csum = 1;
+			desc[curfrag+1].fields.csum_good = 1;
+		}
 
 		if(dma_mapping_error(desc[curfrag+1].fields.address)) {
 			ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
@@ -792,7 +821,11 @@ static int ibmveth_poll(struct net_devic
 			} else {
 				int length = ibmveth_rxq_frame_length(adapter);
 				int offset = ibmveth_rxq_frame_offset(adapter);
+				int csum_good = ibmveth_rxq_csum_good(adapter);
+
 				skb = ibmveth_rxq_get_buffer(adapter);
+				if (csum_good)
+					skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 				ibmveth_rxq_harvest_buffer(adapter);
 
@@ -962,8 +995,10 @@ static void ibmveth_poll_controller(stru
 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 {
 	int rc, i;
+	long ret;
 	struct net_device *netdev;
 	struct ibmveth_adapter *adapter;
+	union ibmveth_illan_attributes set_attr, ret_attr;
 
 	unsigned char *mac_addr_p;
 	unsigned int *mcastFilterSize_p;
@@ -1057,6 +1092,24 @@ static int __devinit ibmveth_probe(struc
 
 	ibmveth_debug_printk("registering netdev...\n");
 
+	ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr.desc);
+
+	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
+	    !ret_attr.fields.trunk_priority &&
+	    ret_attr.fields.csum_offload_padded_pkt_support) {
+		set_attr.desc = 0;
+		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+
+		ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
+					 &ret_attr.desc);
+
+		if (ret == H_SUCCESS)
+			netdev->features |= NETIF_F_IP_CSUM;
+		else
+			ret = h_illan_attributes(dev->unit_address, set_attr.desc,
+						 0, &ret_attr.desc);
+	}
+
 	rc = register_netdev(netdev);
 
 	if(rc) {
diff -puN drivers/net/ibmveth.h~ibmveth_csum_offload drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_csum_offload	2007-08-09 13:08:21.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-09 13:08:21.000000000 -0500
@@ -67,6 +67,21 @@ static inline long h_send_logical_lan(un
 	return rc;
 }
 
+static inline long h_illan_attributes(unsigned long unit_address,
+				      unsigned long reset_mask, unsigned long set_mask,
+				      unsigned long *ret_attributes)
+{
+	long rc;
+	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
+
+	rc = plpar_hcall(H_ILLAN_ATTRIBUTES, retbuf, unit_address,
+			 reset_mask, set_mask);
+
+	*ret_attributes = retbuf[0];
+
+	return rc;
+}
+
 #define h_multicast_ctrl(ua, cmd, mac) \
   plpar_hcall_norets(H_MULTICAST_CTRL, ua, cmd, mac)
 
@@ -141,7 +156,9 @@ struct ibmveth_adapter {
 struct ibmveth_buf_desc_fields {
     u32 valid : 1;
     u32 toggle : 1;
-    u32 reserved : 6;
+    u32 reserved : 4;
+    u32 no_csum : 1;
+    u32 csum_good : 1;
     u32 length : 24;
     u32 address;
 };
@@ -151,10 +168,30 @@ union ibmveth_buf_desc {
     struct ibmveth_buf_desc_fields fields;
 };
 
+struct ibmveth_illan_attributes_fields {
+	u32 reserved;
+	u32 reserved2 : 18;
+	u32 csum_offload_padded_pkt_support : 1;
+	u32 reserved3 : 1;
+	u32 trunk_priority : 4;
+	u32 reserved4 : 5;
+	u32 tcp_csum_offload_ipv6 : 1;
+	u32 tcp_csum_offload_ipv4 : 1;
+	u32 active_trunk : 1;
+};
+
+union ibmveth_illan_attributes {
+	u64 desc;
+	struct ibmveth_illan_attributes_fields fields;
+};
+
 struct ibmveth_rx_q_entry {
     u16 toggle : 1;
     u16 valid : 1;
-    u16 reserved : 14;
+    u16 reserved : 4;
+    u16 no_csum : 1;
+    u16 csum_good : 1;
+    u16 reserved2 : 8;
     u16 offset;
     u32 length;
     u64 correlator;
_

[PATCH 1/6] ibmveth: Enable TCP checksum offload

[Brian King]

This patchset enables TCP checksum offload support for IPV4
on ibmveth. This completely eliminates the generation and checking of
the checksum for packets that are completely virtual and never
touch a physical network. A simple TCP_STREAM netperf run on
a virtual network with maximum mtu set yielded a ~30% increase
in throughput. This feature is enabled by default on systems that
support it, but can be disabled with a module option.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |   53 ++++++++++++++++++++++++++++++++
 linux-2.6-bjking1/drivers/net/ibmveth.h |   41 +++++++++++++++++++++++-
 2 files changed, 92 insertions(+), 2 deletions(-)

diff -puN drivers/net/ibmveth.c~ibmveth_csum_offload drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_csum_offload	2007-08-09 13:08:21.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-09 13:08:21.000000000 -0500
@@ -47,6 +47,8 @@
 #include <linux/mm.h>
 #include <linux/ethtool.h>
 #include <linux/proc_fs.h>
+#include <linux/in.h>
+#include <linux/ip.h>
 #include <asm/semaphore.h>
 #include <asm/hvcall.h>
 #include <asm/atomic.h>
@@ -131,6 +133,11 @@ static inline int ibmveth_rxq_frame_leng
 	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].length);
 }
 
+static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
+{
+	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].csum_good);
+}
+
 /* setup the initial settings for a buffer pool */
 static void ibmveth_init_buffer_pool(struct ibmveth_buff_pool *pool, u32 pool_index, u32 pool_size, u32 buff_size, u32 pool_active)
 {
@@ -684,6 +691,24 @@ static int ibmveth_start_xmit(struct sk_
 					desc[0].fields.length, DMA_TO_DEVICE);
 	desc[0].fields.valid   = 1;
 
+	if (skb->ip_summed == CHECKSUM_PARTIAL &&
+	    ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
+		ibmveth_error_printk("tx: failed to checksum packet\n");
+		tx_dropped++;
+		goto out;
+	}
+
+	if (skb->ip_summed == CHECKSUM_PARTIAL) {
+		unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
+
+		desc[0].fields.no_csum = 1;
+		desc[0].fields.csum_good = 1;
+
+		/* Need to zero out the checksum */
+		buf[0] = 0;
+		buf[1] = 0;
+	}
+
 	if(dma_mapping_error(desc[0].fields.address)) {
 		ibmveth_error_printk("tx: unable to map initial fragment\n");
 		tx_map_failed++;
@@ -702,6 +727,10 @@ static int ibmveth_start_xmit(struct sk_
 				frag->size, DMA_TO_DEVICE);
 		desc[curfrag+1].fields.length = frag->size;
 		desc[curfrag+1].fields.valid  = 1;
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			desc[curfrag+1].fields.no_csum = 1;
+			desc[curfrag+1].fields.csum_good = 1;
+		}
 
 		if(dma_mapping_error(desc[curfrag+1].fields.address)) {
 			ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
@@ -792,7 +821,11 @@ static int ibmveth_poll(struct net_devic
 			} else {
 				int length = ibmveth_rxq_frame_length(adapter);
 				int offset = ibmveth_rxq_frame_offset(adapter);
+				int csum_good = ibmveth_rxq_csum_good(adapter);
+
 				skb = ibmveth_rxq_get_buffer(adapter);
+				if (csum_good)
+					skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 				ibmveth_rxq_harvest_buffer(adapter);
 
@@ -962,8 +995,10 @@ static void ibmveth_poll_controller(stru
 static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 {
 	int rc, i;
+	long ret;
 	struct net_device *netdev;
 	struct ibmveth_adapter *adapter;
+	union ibmveth_illan_attributes set_attr, ret_attr;
 
 	unsigned char *mac_addr_p;
 	unsigned int *mcastFilterSize_p;
@@ -1057,6 +1092,24 @@ static int __devinit ibmveth_probe(struc
 
 	ibmveth_debug_printk("registering netdev...\n");
 
+	ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr.desc);
+
+	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
+	    !ret_attr.fields.trunk_priority &&
+	    ret_attr.fields.csum_offload_padded_pkt_support) {
+		set_attr.desc = 0;
+		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+
+		ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
+					 &ret_attr.desc);
+
+		if (ret == H_SUCCESS)
+			netdev->features |= NETIF_F_IP_CSUM;
+		else
+			ret = h_illan_attributes(dev->unit_address, set_attr.desc,
+						 0, &ret_attr.desc);
+	}
+
 	rc = register_netdev(netdev);
 
 	if(rc) {
diff -puN drivers/net/ibmveth.h~ibmveth_csum_offload drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_csum_offload	2007-08-09 13:08:21.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-09 13:08:21.000000000 -0500
@@ -67,6 +67,21 @@ static inline long h_send_logical_lan(un
 	return rc;
 }
 
+static inline long h_illan_attributes(unsigned long unit_address,
+				      unsigned long reset_mask, unsigned long set_mask,
+				      unsigned long *ret_attributes)
+{
+	long rc;
+	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
+
+	rc = plpar_hcall(H_ILLAN_ATTRIBUTES, retbuf, unit_address,
+			 reset_mask, set_mask);
+
+	*ret_attributes = retbuf[0];
+
+	return rc;
+}
+
 #define h_multicast_ctrl(ua, cmd, mac) \
   plpar_hcall_norets(H_MULTICAST_CTRL, ua, cmd, mac)
 
@@ -141,7 +156,9 @@ struct ibmveth_adapter {
 struct ibmveth_buf_desc_fields {
     u32 valid : 1;
     u32 toggle : 1;
-    u32 reserved : 6;
+    u32 reserved : 4;
+    u32 no_csum : 1;
+    u32 csum_good : 1;
     u32 length : 24;
     u32 address;
 };
@@ -151,10 +168,30 @@ union ibmveth_buf_desc {
     struct ibmveth_buf_desc_fields fields;
 };
 
+struct ibmveth_illan_attributes_fields {
+	u32 reserved;
+	u32 reserved2 : 18;
+	u32 csum_offload_padded_pkt_support : 1;
+	u32 reserved3 : 1;
+	u32 trunk_priority : 4;
+	u32 reserved4 : 5;
+	u32 tcp_csum_offload_ipv6 : 1;
+	u32 tcp_csum_offload_ipv4 : 1;
+	u32 active_trunk : 1;
+};
+
+union ibmveth_illan_attributes {
+	u64 desc;
+	struct ibmveth_illan_attributes_fields fields;
+};
+
 struct ibmveth_rx_q_entry {
     u16 toggle : 1;
     u16 valid : 1;
-    u16 reserved : 14;
+    u16 reserved : 4;
+    u16 no_csum : 1;
+    u16 csum_good : 1;
+    u16 reserved2 : 8;
     u16 offset;
     u32 length;
     u64 correlator;
_

[PATCH 2/6] ibmveth: Implement ethtool hooks to enable/disable checksum offload

[Brian King]

This patch adds the appropriate ethtool hooks to allow for enabling/disabling
of hypervisor assisted checksum offload for TCP.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |  125 +++++++++++++++++++++++++++++++-
 linux-2.6-bjking1/drivers/net/ibmveth.h |    1 
 2 files changed, 124 insertions(+), 2 deletions(-)

diff -puN drivers/net/ibmveth.c~ibmveth_csum_offload_ethtool drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_csum_offload_ethtool	2007-08-09 15:15:07.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-09 15:15:07.000000000 -0500
@@ -641,12 +641,132 @@ static u32 netdev_get_link(struct net_de
 	return 1;
 }
 
+static void ibmveth_set_rx_csum_flags(struct net_device *dev, u32 data)
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+
+	if (data)
+		adapter->rx_csum = 1;
+	else {
+		/*
+		 * Since the ibmveth firmware interface does not have the concept of
+		 * separate tx/rx checksum offload enable, if rx checksum is disabled
+		 * we also have to disable tx checksum offload. Once we disable rx
+		 * checksum offload, we are no longer allowed to send tx buffers that
+		 * are not properly checksummed.
+		 */
+		adapter->rx_csum = 0;
+		dev->features &= ~NETIF_F_IP_CSUM;
+	}
+}
+
+static void ibmveth_set_tx_csum_flags(struct net_device *dev, u32 data)
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+
+	if (data) {
+		dev->features |= NETIF_F_IP_CSUM;
+		adapter->rx_csum = 1;
+	} else
+		dev->features &= ~NETIF_F_IP_CSUM;
+}
+
+static int ibmveth_set_csum_offload(struct net_device *dev, u32 data,
+				    void (*done) (struct net_device *, u32))
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+	union ibmveth_illan_attributes set_attr, clr_attr, ret_attr;
+	long ret;
+	int rc1 = 0, rc2 = 0;
+	int restart = 0;
+
+	if (netif_running(dev)) {
+		restart = 1;
+		adapter->pool_config = 1;
+		ibmveth_close(dev);
+		adapter->pool_config = 0;
+	}
+
+	set_attr.desc = 0;
+	clr_attr.desc = 0;
+
+	if (data)
+		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+	else
+		clr_attr.fields.tcp_csum_offload_ipv4 = 1;
+
+	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr.desc);
+
+	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
+	    !ret_attr.fields.trunk_priority &&
+	    ret_attr.fields.csum_offload_padded_pkt_support) {
+		ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr.desc,
+					 set_attr.desc, &ret_attr.desc);
+
+		if (ret != H_SUCCESS) {
+			rc1 = -EIO;
+			ibmveth_error_printk("unable to change checksum offload settings."
+					     " %d rc=%ld\n", data, ret);
+
+			ret = h_illan_attributes(adapter->vdev->unit_address,
+						 set_attr.desc, clr_attr.desc, &ret_attr.desc);
+		} else
+			done(dev, data);
+	} else {
+		rc1 = -EIO;
+		ibmveth_error_printk("unable to change checksum offload settings."
+				     " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr.desc);
+	}
+
+	if (restart)
+		rc2 = ibmveth_open(dev);
+
+	return rc1 ? rc1 : rc2;
+}
+
+static int ibmveth_set_rx_csum(struct net_device *dev, u32 data)
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+
+	if ((data && adapter->rx_csum) || (!data && !adapter->rx_csum))
+		return 0;
+
+	return ibmveth_set_csum_offload(dev, data, ibmveth_set_rx_csum_flags);
+}
+
+static int ibmveth_set_tx_csum(struct net_device *dev, u32 data)
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+	int rc = 0;
+
+	if (data && (dev->features & NETIF_F_IP_CSUM))
+		return 0;
+	if (!data && !(dev->features & NETIF_F_IP_CSUM))
+		return 0;
+
+	if (data && !adapter->rx_csum)
+		rc = ibmveth_set_csum_offload(dev, data, ibmveth_set_tx_csum_flags);
+	else
+		ibmveth_set_tx_csum_flags(dev, data);
+
+	return rc;
+}
+
+static u32 ibmveth_get_rx_csum(struct net_device *dev)
+{
+	struct ibmveth_adapter *adapter = dev->priv;
+	return adapter->rx_csum;
+}
+
 static const struct ethtool_ops netdev_ethtool_ops = {
 	.get_drvinfo		= netdev_get_drvinfo,
 	.get_settings		= netdev_get_settings,
 	.get_link		= netdev_get_link,
 	.get_sg			= ethtool_op_get_sg,
 	.get_tx_csum		= ethtool_op_get_tx_csum,
+	.set_tx_csum		= ibmveth_set_tx_csum,
+	.get_rx_csum		= ibmveth_get_rx_csum,
+	.set_rx_csum		= ibmveth_set_rx_csum,
 };
 
 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
@@ -1103,9 +1223,10 @@ static int __devinit ibmveth_probe(struc
 		ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
 					 &ret_attr.desc);
 
-		if (ret == H_SUCCESS)
+		if (ret == H_SUCCESS) {
+			adapter->rx_csum = 1;
 			netdev->features |= NETIF_F_IP_CSUM;
-		else
+		} else
 			ret = h_illan_attributes(dev->unit_address, set_attr.desc,
 						 0, &ret_attr.desc);
 	}
diff -puN drivers/net/ibmveth.h~ibmveth_csum_offload_ethtool drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_csum_offload_ethtool	2007-08-09 15:15:07.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-09 15:15:07.000000000 -0500
@@ -137,6 +137,7 @@ struct ibmveth_adapter {
     struct ibmveth_buff_pool rx_buff_pool[IbmVethNumBufferPools];
     struct ibmveth_rx_q rx_queue;
     int pool_config;
+    int rx_csum;
 
     /* adapter specific stats */
     u64 replenish_task_cycles;
_

[PATCH 3/6] ibmveth: Add ethtool TSO handlers

[Brian King]

Add handlers for get_tso and get_ufo to prevent errors being printed
by ethtool.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |    2 ++
 1 file changed, 2 insertions(+)

diff -puN drivers/net/ibmveth.c~ibmveth_ethtool_get_tso drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_ethtool_get_tso	2007-08-08 10:46:28.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-08 10:46:28.000000000 -0500
@@ -767,6 +767,8 @@ static const struct ethtool_ops netdev_e
 	.set_tx_csum		= ibmveth_set_tx_csum,
 	.get_rx_csum		= ibmveth_get_rx_csum,
 	.set_rx_csum		= ibmveth_set_rx_csum,
+	.get_tso			= ethtool_op_get_tso,
+	.get_ufo			= ethtool_op_get_ufo,
 };
 
 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
_

[PATCH 4/6] ibmveth: Add ethtool driver stats hooks

[Brian King]

Add ethtool hooks to ibmveth to retrieve driver statistics.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |   51 ++++++++++++++++++++++++++++++++
 1 file changed, 51 insertions(+)

diff -puN drivers/net/ibmveth.c~ibmveth_ethtool_driver_stats drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_ethtool_driver_stats	2007-08-08 10:46:30.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-08 10:48:08.000000000 -0500
@@ -112,6 +112,28 @@ MODULE_DESCRIPTION("IBM i/pSeries Virtua
 MODULE_LICENSE("GPL");
 MODULE_VERSION(ibmveth_driver_version);
 
+struct ibmveth_stat {
+	char name[ETH_GSTRING_LEN];
+	int offset;
+};
+
+#define IBMVETH_STAT_OFF(stat) offsetof(struct ibmveth_adapter, stat)
+#define IBMVETH_GET_STAT(a, off) *((u64 *)(((unsigned long)(a)) + off))
+
+struct ibmveth_stat ibmveth_stats[] = {
+	{ "replenish_task_cycles", IBMVETH_STAT_OFF(replenish_task_cycles) },
+	{ "replenish_no_mem", IBMVETH_STAT_OFF(replenish_no_mem) },
+	{ "replenish_add_buff_failure", IBMVETH_STAT_OFF(replenish_add_buff_failure) },
+	{ "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
+	{ "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
+	{ "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
+	{ "tx_multidesc_send", IBMVETH_STAT_OFF(tx_multidesc_send) },
+	{ "tx_linearized", IBMVETH_STAT_OFF(tx_linearized) },
+	{ "tx_linearize_failed", IBMVETH_STAT_OFF(tx_linearize_failed) },
+	{ "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
+	{ "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
+};
+
 /* simple methods of getting data from the current rxq entry */
 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 {
@@ -758,6 +780,32 @@ static u32 ibmveth_get_rx_csum(struct ne
 	return adapter->rx_csum;
 }
 
+static void ibmveth_get_strings(struct net_device *dev, u32 stringset, u8 *data)
+{
+	int i;
+
+	if (stringset != ETH_SS_STATS)
+		return;
+
+	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++, data += ETH_GSTRING_LEN)
+		memcpy(data, ibmveth_stats[i].name, ETH_GSTRING_LEN);
+}
+
+static int ibmveth_get_stats_count(struct net_device *dev)
+{
+	return ARRAY_SIZE(ibmveth_stats);
+}
+
+static void ibmveth_get_ethtool_stats(struct net_device *dev,
+				      struct ethtool_stats *stats, u64 *data)
+{
+	int i;
+	struct ibmveth_adapter *adapter = dev->priv;
+
+	for (i = 0; i < ARRAY_SIZE(ibmveth_stats); i++)
+		data[i] = IBMVETH_GET_STAT(adapter, ibmveth_stats[i].offset);
+}
+
 static const struct ethtool_ops netdev_ethtool_ops = {
 	.get_drvinfo		= netdev_get_drvinfo,
 	.get_settings		= netdev_get_settings,
@@ -769,6 +817,9 @@ static const struct ethtool_ops netdev_e
 	.set_rx_csum		= ibmveth_set_rx_csum,
 	.get_tso			= ethtool_op_get_tso,
 	.get_ufo			= ethtool_op_get_ufo,
+	.get_strings		= ibmveth_get_strings,
+	.get_stats_count		= ibmveth_get_stats_count,
+	.get_ethtool_stats	= ibmveth_get_ethtool_stats,
 };
 
 static int ibmveth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
diff -puN drivers/net/ibmveth.h~ibmveth_ethtool_driver_stats drivers/net/ibmveth.h
_

[PATCH 5/6] ibmveth: Remove dead frag processing code

[Brian King]

Removes dead frag processing code from ibmveth. Since NETIF_F_SG was
not set, this code was never executed. Also, since the ibmveth
interface can only handle 6 fragments, core networking code would need
to be modified in order to efficiently enable this support.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |  100 +++++---------------------------
 linux-2.6-bjking1/drivers/net/ibmveth.h |    5 -
 2 files changed, 17 insertions(+), 88 deletions(-)

diff -puN drivers/net/ibmveth.c~ibmveth_remove_frag drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_remove_frag	2007-08-09 15:15:18.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-09 15:15:18.000000000 -0500
@@ -28,7 +28,6 @@
 /**************************************************************************/
 /*
   TODO:
-  - remove frag processing code - no longer needed
   - add support for sysfs
   - possibly remove procfs support
 */
@@ -127,9 +126,6 @@ struct ibmveth_stat ibmveth_stats[] = {
 	{ "replenish_add_buff_success", IBMVETH_STAT_OFF(replenish_add_buff_success) },
 	{ "rx_invalid_buffer", IBMVETH_STAT_OFF(rx_invalid_buffer) },
 	{ "rx_no_buffer", IBMVETH_STAT_OFF(rx_no_buffer) },
-	{ "tx_multidesc_send", IBMVETH_STAT_OFF(tx_multidesc_send) },
-	{ "tx_linearized", IBMVETH_STAT_OFF(tx_linearized) },
-	{ "tx_linearize_failed", IBMVETH_STAT_OFF(tx_linearize_failed) },
 	{ "tx_map_failed", IBMVETH_STAT_OFF(tx_map_failed) },
 	{ "tx_send_failed", IBMVETH_STAT_OFF(tx_send_failed) },
 };
@@ -832,9 +828,8 @@ static int ibmveth_ioctl(struct net_devi
 static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 {
 	struct ibmveth_adapter *adapter = netdev->priv;
-	union ibmveth_buf_desc desc[IbmVethMaxSendFrags];
+	union ibmveth_buf_desc desc;
 	unsigned long lpar_rc;
-	int nfrags = 0, curfrag;
 	unsigned long correlator;
 	unsigned long flags;
 	unsigned int retry_count;
@@ -844,25 +839,11 @@ static int ibmveth_start_xmit(struct sk_
 	unsigned int tx_send_failed = 0;
 	unsigned int tx_map_failed = 0;
 
-
-	if ((skb_shinfo(skb)->nr_frags + 1) > IbmVethMaxSendFrags) {
-		tx_dropped++;
-		goto out;
-	}
-
-	memset(&desc, 0, sizeof(desc));
-
-	/* nfrags = number of frags after the initial fragment */
-	nfrags = skb_shinfo(skb)->nr_frags;
-
-	if(nfrags)
-		adapter->tx_multidesc_send++;
-
-	/* map the initial fragment */
-	desc[0].fields.length  = nfrags ? skb->len - skb->data_len : skb->len;
-	desc[0].fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
-					desc[0].fields.length, DMA_TO_DEVICE);
-	desc[0].fields.valid   = 1;
+	desc.desc = 0;
+	desc.fields.length  = skb->len;
+	desc.fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
+					     desc.fields.length, DMA_TO_DEVICE);
+	desc.fields.valid   = 1;
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 	    ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
@@ -874,75 +855,34 @@ static int ibmveth_start_xmit(struct sk_
 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 		unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
 
-		desc[0].fields.no_csum = 1;
-		desc[0].fields.csum_good = 1;
+		desc.fields.no_csum = 1;
+		desc.fields.csum_good = 1;
 
 		/* Need to zero out the checksum */
 		buf[0] = 0;
 		buf[1] = 0;
 	}
 
-	if(dma_mapping_error(desc[0].fields.address)) {
-		ibmveth_error_printk("tx: unable to map initial fragment\n");
+	if (dma_mapping_error(desc.fields.address)) {
+		ibmveth_error_printk("tx: unable to map xmit buffer\n");
 		tx_map_failed++;
 		tx_dropped++;
 		goto out;
 	}
 
-	curfrag = nfrags;
-
-	/* map fragments past the initial portion if there are any */
-	while(curfrag--) {
-		skb_frag_t *frag = &skb_shinfo(skb)->frags[curfrag];
-		desc[curfrag+1].fields.address
-			= dma_map_single(&adapter->vdev->dev,
-				page_address(frag->page) + frag->page_offset,
-				frag->size, DMA_TO_DEVICE);
-		desc[curfrag+1].fields.length = frag->size;
-		desc[curfrag+1].fields.valid  = 1;
-		if (skb->ip_summed == CHECKSUM_PARTIAL) {
-			desc[curfrag+1].fields.no_csum = 1;
-			desc[curfrag+1].fields.csum_good = 1;
-		}
-
-		if(dma_mapping_error(desc[curfrag+1].fields.address)) {
-			ibmveth_error_printk("tx: unable to map fragment %d\n", curfrag);
-			tx_map_failed++;
-			tx_dropped++;
-			/* Free all the mappings we just created */
-			while(curfrag < nfrags) {
-				dma_unmap_single(&adapter->vdev->dev,
-						 desc[curfrag+1].fields.address,
-						 desc[curfrag+1].fields.length,
-						 DMA_TO_DEVICE);
-				curfrag++;
-			}
-			goto out;
-		}
-	}
-
 	/* send the frame. Arbitrarily set retrycount to 1024 */
 	correlator = 0;
 	retry_count = 1024;
 	do {
 		lpar_rc = h_send_logical_lan(adapter->vdev->unit_address,
-					     desc[0].desc,
-					     desc[1].desc,
-					     desc[2].desc,
-					     desc[3].desc,
-					     desc[4].desc,
-					     desc[5].desc,
-					     correlator,
-					     &correlator);
+					     desc.desc, 0, 0, 0, 0, 0,
+					     correlator, &correlator);
 	} while ((lpar_rc == H_BUSY) && (retry_count--));
 
 	if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
-		int i;
 		ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
-		for(i = 0; i < 6; i++) {
-			ibmveth_error_printk("tx: desc[%i] valid=%d, len=%d, address=0x%d\n", i,
-					     desc[i].fields.valid, desc[i].fields.length, desc[i].fields.address);
-		}
+		ibmveth_error_printk("tx: valid=%d, len=%d, address=0x%08x\n",
+				     desc.fields.valid, desc.fields.length, desc.fields.address);
 		tx_send_failed++;
 		tx_dropped++;
 	} else {
@@ -951,11 +891,8 @@ static int ibmveth_start_xmit(struct sk_
 		netdev->trans_start = jiffies;
 	}
 
-	do {
-		dma_unmap_single(&adapter->vdev->dev,
-				desc[nfrags].fields.address,
-				desc[nfrags].fields.length, DMA_TO_DEVICE);
-	} while(--nfrags >= 0);
+	dma_unmap_single(&adapter->vdev->dev, desc.fields.address,
+			 desc.fields.length, DMA_TO_DEVICE);
 
 out:	spin_lock_irqsave(&adapter->stats_lock, flags);
 	adapter->stats.tx_dropped += tx_dropped;
@@ -1366,10 +1303,7 @@ static int ibmveth_seq_show(struct seq_f
 		   firmware_mac[3], firmware_mac[4], firmware_mac[5]);
 
 	seq_printf(seq, "\nAdapter Statistics:\n");
-	seq_printf(seq, "  TX:  skbuffs linearized:          %ld\n", adapter->tx_linearized);
-	seq_printf(seq, "       multi-descriptor sends:      %ld\n", adapter->tx_multidesc_send);
-	seq_printf(seq, "       skb_linearize failures:      %ld\n", adapter->tx_linearize_failed);
-	seq_printf(seq, "       vio_map_single failres:      %ld\n", adapter->tx_map_failed);
+	seq_printf(seq, "  TX:  vio_map_single failres:      %ld\n", adapter->tx_map_failed);
 	seq_printf(seq, "       send failures:               %ld\n", adapter->tx_send_failed);
 	seq_printf(seq, "  RX:  replenish task cycles:       %ld\n", adapter->replenish_task_cycles);
 	seq_printf(seq, "       alloc_skb_failures:          %ld\n", adapter->replenish_no_mem);
diff -puN drivers/net/ibmveth.h~ibmveth_remove_frag drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_remove_frag	2007-08-09 15:15:18.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-09 15:15:18.000000000 -0500
@@ -25,8 +25,6 @@
 #ifndef _IBMVETH_H
 #define _IBMVETH_H
 
-#define IbmVethMaxSendFrags 6
-
 /* constants for H_MULTICAST_CTRL */
 #define IbmVethMcastReceptionModifyBit     0x80000UL
 #define IbmVethMcastReceptionEnableBit     0x20000UL
@@ -146,9 +144,6 @@ struct ibmveth_adapter {
     u64 replenish_add_buff_success;
     u64 rx_invalid_buffer;
     u64 rx_no_buffer;
-    u64 tx_multidesc_send;
-    u64 tx_linearized;
-    u64 tx_linearize_failed;
     u64 tx_map_failed;
     u64 tx_send_failed;
     spinlock_t stats_lock;
_

[PATCH 6/6] ibmveth: Remove use of bitfields

[Brian King]

Removes the use of bitfields from the ibmveth driver. This results
in slightly smaller object code.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
---

 linux-2.6-bjking1/drivers/net/ibmveth.c |   90 ++++++++++++++++----------------
 linux-2.6-bjking1/drivers/net/ibmveth.h |   56 ++++++++-----------
 2 files changed, 68 insertions(+), 78 deletions(-)

diff -puN drivers/net/ibmveth.h~ibmveth_nobitfields drivers/net/ibmveth.h
--- linux-2.6/drivers/net/ibmveth.h~ibmveth_nobitfields	2007-08-09 15:15:27.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.h	2007-08-09 15:15:27.000000000 -0500
@@ -39,6 +39,12 @@
 #define IbmVethMcastRemoveFilter     0x2UL
 #define IbmVethMcastClearFilterTable 0x3UL
 
+#define IBMVETH_ILLAN_PADDED_PKT_CSUM	0x0000000000002000ULL
+#define IBMVETH_ILLAN_TRUNK_PRI_MASK	0x0000000000000F00ULL
+#define IBMVETH_ILLAN_IPV6_TCP_CSUM		0x0000000000000004ULL
+#define IBMVETH_ILLAN_IPV4_TCP_CSUM		0x0000000000000002ULL
+#define IBMVETH_ILLAN_ACTIVE_TRUNK		0x0000000000000001ULL
+
 /* hcall macros */
 #define h_register_logical_lan(ua, buflst, rxq, fltlst, mac) \
   plpar_hcall_norets(H_REGISTER_LOGICAL_LAN, ua, buflst, rxq, fltlst, mac)
@@ -150,13 +156,13 @@ struct ibmveth_adapter {
 };
 
 struct ibmveth_buf_desc_fields {
-    u32 valid : 1;
-    u32 toggle : 1;
-    u32 reserved : 4;
-    u32 no_csum : 1;
-    u32 csum_good : 1;
-    u32 length : 24;
-    u32 address;
+	u32 flags_len;
+#define IBMVETH_BUF_VALID	0x80000000
+#define IBMVETH_BUF_TOGGLE	0x40000000
+#define IBMVETH_BUF_NO_CSUM	0x02000000
+#define IBMVETH_BUF_CSUM_GOOD	0x01000000
+#define IBMVETH_BUF_LEN_MASK	0x00FFFFFF
+	u32 address;
 };
 
 union ibmveth_buf_desc {
@@ -164,33 +170,17 @@ union ibmveth_buf_desc {
     struct ibmveth_buf_desc_fields fields;
 };
 
-struct ibmveth_illan_attributes_fields {
-	u32 reserved;
-	u32 reserved2 : 18;
-	u32 csum_offload_padded_pkt_support : 1;
-	u32 reserved3 : 1;
-	u32 trunk_priority : 4;
-	u32 reserved4 : 5;
-	u32 tcp_csum_offload_ipv6 : 1;
-	u32 tcp_csum_offload_ipv4 : 1;
-	u32 active_trunk : 1;
-};
-
-union ibmveth_illan_attributes {
-	u64 desc;
-	struct ibmveth_illan_attributes_fields fields;
-};
-
 struct ibmveth_rx_q_entry {
-    u16 toggle : 1;
-    u16 valid : 1;
-    u16 reserved : 4;
-    u16 no_csum : 1;
-    u16 csum_good : 1;
-    u16 reserved2 : 8;
-    u16 offset;
-    u32 length;
-    u64 correlator;
+	u32 flags_off;
+#define IBMVETH_RXQ_TOGGLE		0x80000000
+#define IBMVETH_RXQ_TOGGLE_SHIFT	31
+#define IBMVETH_RXQ_VALID		0x40000000
+#define IBMVETH_RXQ_NO_CSUM		0x02000000
+#define IBMVETH_RXQ_CSUM_GOOD		0x01000000
+#define IBMVETH_RXQ_OFF_MASK		0x0000FFFF
+
+	u32 length;
+	u64 correlator;
 };
 
 #endif /* _IBMVETH_H */
diff -puN drivers/net/ibmveth.c~ibmveth_nobitfields drivers/net/ibmveth.c
--- linux-2.6/drivers/net/ibmveth.c~ibmveth_nobitfields	2007-08-09 15:15:27.000000000 -0500
+++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-09 15:15:27.000000000 -0500
@@ -131,19 +131,29 @@ struct ibmveth_stat ibmveth_stats[] = {
 };
 
 /* simple methods of getting data from the current rxq entry */
+static inline u32 ibmveth_rxq_flags(struct ibmveth_adapter *adapter)
+{
+	return adapter->rx_queue.queue_addr[adapter->rx_queue.index].flags_off;
+}
+
+static inline int ibmveth_rxq_toggle(struct ibmveth_adapter *adapter)
+{
+	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_TOGGLE) >> IBMVETH_RXQ_TOGGLE_SHIFT;
+}
+
 static inline int ibmveth_rxq_pending_buffer(struct ibmveth_adapter *adapter)
 {
-	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].toggle == adapter->rx_queue.toggle);
+	return (ibmveth_rxq_toggle(adapter) == adapter->rx_queue.toggle);
 }
 
 static inline int ibmveth_rxq_buffer_valid(struct ibmveth_adapter *adapter)
 {
-	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].valid);
+	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_VALID);
 }
 
 static inline int ibmveth_rxq_frame_offset(struct ibmveth_adapter *adapter)
 {
-	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].offset);
+	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_OFF_MASK);
 }
 
 static inline int ibmveth_rxq_frame_length(struct ibmveth_adapter *adapter)
@@ -153,7 +163,7 @@ static inline int ibmveth_rxq_frame_leng
 
 static inline int ibmveth_rxq_csum_good(struct ibmveth_adapter *adapter)
 {
-	return (adapter->rx_queue.queue_addr[adapter->rx_queue.index].csum_good);
+	return (ibmveth_rxq_flags(adapter) & IBMVETH_RXQ_CSUM_GOOD);
 }
 
 /* setup the initial settings for a buffer pool */
@@ -253,9 +263,7 @@ static void ibmveth_replenish_buffer_poo
 		correlator = ((u64)pool->index << 32) | index;
 		*(u64*)skb->data = correlator;
 
-		desc.desc = 0;
-		desc.fields.valid = 1;
-		desc.fields.length = pool->buff_size;
+		desc.fields.flags_len = IBMVETH_BUF_VALID | pool->buff_size;
 		desc.fields.address = dma_addr;
 
 		lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
@@ -396,9 +404,8 @@ static void ibmveth_rxq_recycle_buffer(s
 		return;
 	}
 
-	desc.desc = 0;
-	desc.fields.valid = 1;
-	desc.fields.length = adapter->rx_buff_pool[pool].buff_size;
+	desc.fields.flags_len = IBMVETH_BUF_VALID |
+		adapter->rx_buff_pool[pool].buff_size;
 	desc.fields.address = adapter->rx_buff_pool[pool].dma_addr[index];
 
 	lpar_rc = h_add_logical_lan_buffer(adapter->vdev->unit_address, desc.desc);
@@ -549,9 +556,7 @@ static int ibmveth_open(struct net_devic
 	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
 	mac_address = mac_address >> 16;
 
-	rxq_desc.desc = 0;
-	rxq_desc.fields.valid = 1;
-	rxq_desc.fields.length = adapter->rx_queue.queue_len;
+	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID | adapter->rx_queue.queue_len;
 	rxq_desc.fields.address = adapter->rx_queue.queue_dma;
 
 	ibmveth_debug_printk("buffer list @ 0x%p\n", adapter->buffer_list_addr);
@@ -693,7 +698,7 @@ static int ibmveth_set_csum_offload(stru
 				    void (*done) (struct net_device *, u32))
 {
 	struct ibmveth_adapter *adapter = dev->priv;
-	union ibmveth_illan_attributes set_attr, clr_attr, ret_attr;
+	u64 set_attr, clr_attr, ret_attr;
 	long ret;
 	int rc1 = 0, rc2 = 0;
 	int restart = 0;
@@ -705,21 +710,21 @@ static int ibmveth_set_csum_offload(stru
 		adapter->pool_config = 0;
 	}
 
-	set_attr.desc = 0;
-	clr_attr.desc = 0;
+	set_attr = 0;
+	clr_attr = 0;
 
 	if (data)
-		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 	else
-		clr_attr.fields.tcp_csum_offload_ipv4 = 1;
+		clr_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 
-	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr.desc);
+	ret = h_illan_attributes(adapter->vdev->unit_address, 0, 0, &ret_attr);
 
-	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
-	    !ret_attr.fields.trunk_priority &&
-	    ret_attr.fields.csum_offload_padded_pkt_support) {
-		ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr.desc,
-					 set_attr.desc, &ret_attr.desc);
+	if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
+	    !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
+	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
+		ret = h_illan_attributes(adapter->vdev->unit_address, clr_attr,
+					 set_attr, &ret_attr);
 
 		if (ret != H_SUCCESS) {
 			rc1 = -EIO;
@@ -727,13 +732,13 @@ static int ibmveth_set_csum_offload(stru
 					     " %d rc=%ld\n", data, ret);
 
 			ret = h_illan_attributes(adapter->vdev->unit_address,
-						 set_attr.desc, clr_attr.desc, &ret_attr.desc);
+						 set_attr, clr_attr, &ret_attr);
 		} else
 			done(dev, data);
 	} else {
 		rc1 = -EIO;
 		ibmveth_error_printk("unable to change checksum offload settings."
-				     " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr.desc);
+				     " %d rc=%ld ret_attr=%lx\n", data, ret, ret_attr);
 	}
 
 	if (restart)
@@ -839,11 +844,9 @@ static int ibmveth_start_xmit(struct sk_
 	unsigned int tx_send_failed = 0;
 	unsigned int tx_map_failed = 0;
 
-	desc.desc = 0;
-	desc.fields.length  = skb->len;
+	desc.fields.flags_len = IBMVETH_BUF_VALID | skb->len;
 	desc.fields.address = dma_map_single(&adapter->vdev->dev, skb->data,
-					     desc.fields.length, DMA_TO_DEVICE);
-	desc.fields.valid   = 1;
+					     skb->len, DMA_TO_DEVICE);
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 	    ip_hdr(skb)->protocol != IPPROTO_TCP && skb_checksum_help(skb)) {
@@ -855,8 +858,7 @@ static int ibmveth_start_xmit(struct sk_
 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 		unsigned char *buf = skb_transport_header(skb) + skb->csum_offset;
 
-		desc.fields.no_csum = 1;
-		desc.fields.csum_good = 1;
+		desc.fields.flags_len |= (IBMVETH_BUF_NO_CSUM | IBMVETH_BUF_CSUM_GOOD);
 
 		/* Need to zero out the checksum */
 		buf[0] = 0;
@@ -882,7 +884,8 @@ static int ibmveth_start_xmit(struct sk_
 	if(lpar_rc != H_SUCCESS && lpar_rc != H_DROPPED) {
 		ibmveth_error_printk("tx: h_send_logical_lan failed with rc=%ld\n", lpar_rc);
 		ibmveth_error_printk("tx: valid=%d, len=%d, address=0x%08x\n",
-				     desc.fields.valid, desc.fields.length, desc.fields.address);
+				     (desc.fields.flags_len & IBMVETH_BUF_VALID) ? 1 : 0,
+				     skb->len, desc.fields.address);
 		tx_send_failed++;
 		tx_dropped++;
 	} else {
@@ -892,7 +895,7 @@ static int ibmveth_start_xmit(struct sk_
 	}
 
 	dma_unmap_single(&adapter->vdev->dev, desc.fields.address,
-			 desc.fields.length, DMA_TO_DEVICE);
+			 skb->len, DMA_TO_DEVICE);
 
 out:	spin_lock_irqsave(&adapter->stats_lock, flags);
 	adapter->stats.tx_dropped += tx_dropped;
@@ -1108,7 +1111,7 @@ static int __devinit ibmveth_probe(struc
 	long ret;
 	struct net_device *netdev;
 	struct ibmveth_adapter *adapter;
-	union ibmveth_illan_attributes set_attr, ret_attr;
+	u64 set_attr, ret_attr;
 
 	unsigned char *mac_addr_p;
 	unsigned int *mcastFilterSize_p;
@@ -1202,23 +1205,20 @@ static int __devinit ibmveth_probe(struc
 
 	ibmveth_debug_printk("registering netdev...\n");
 
-	ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr.desc);
+	ret = h_illan_attributes(dev->unit_address, 0, 0, &ret_attr);
 
-	if (ret == H_SUCCESS && !ret_attr.fields.active_trunk &&
-	    !ret_attr.fields.trunk_priority &&
-	    ret_attr.fields.csum_offload_padded_pkt_support) {
-		set_attr.desc = 0;
-		set_attr.fields.tcp_csum_offload_ipv4 = 1;
+	if (ret == H_SUCCESS && !(ret_attr & IBMVETH_ILLAN_ACTIVE_TRUNK) &&
+	    !(ret_attr & IBMVETH_ILLAN_TRUNK_PRI_MASK) &&
+	    (ret_attr & IBMVETH_ILLAN_PADDED_PKT_CSUM)) {
+		set_attr = IBMVETH_ILLAN_IPV4_TCP_CSUM;
 
-		ret = h_illan_attributes(dev->unit_address, 0, set_attr.desc,
-					 &ret_attr.desc);
+		ret = h_illan_attributes(dev->unit_address, 0, set_attr, &ret_attr);
 
 		if (ret == H_SUCCESS) {
 			adapter->rx_csum = 1;
 			netdev->features |= NETIF_F_IP_CSUM;
 		} else
-			ret = h_illan_attributes(dev->unit_address, set_attr.desc,
-						 0, &ret_attr.desc);
+			ret = h_illan_attributes(dev->unit_address, set_attr, 0, &ret_attr);
 	}
 
 	rc = register_netdev(netdev);
_

Re: [PATCH 3/6] ibmveth: Add ethtool TSO handlers

[Jeff Garzik]

Brian King wrote:
> Add handlers for get_tso and get_ufo to prevent errors being printed
> by ethtool.
> 
> Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
> ---
> 
>  linux-2.6-bjking1/drivers/net/ibmveth.c |    2 ++
>  1 file changed, 2 insertions(+)
> 
> diff -puN drivers/net/ibmveth.c~ibmveth_ethtool_get_tso drivers/net/ibmveth.c
> --- linux-2.6/drivers/net/ibmveth.c~ibmveth_ethtool_get_tso	2007-08-08 10:46:28.000000000 -0500
> +++ linux-2.6-bjking1/drivers/net/ibmveth.c	2007-08-08 10:46:28.000000000 -0500
> @@ -767,6 +767,8 @@ static const struct ethtool_ops netdev_e
>  	.set_tx_csum		= ibmveth_set_tx_csum,
>  	.get_rx_csum		= ibmveth_get_rx_csum,
>  	.set_rx_csum		= ibmveth_set_rx_csum,
> +	.get_tso			= ethtool_op_get_tso,
> +	.get_ufo			= ethtool_op_get_ufo,

This patch is fine, but I wonder if we shouldn't add some code to 
net/core/ethtool.c along the lines of...

	if (!netdev->ethtool_ops->get_tso)
		ethtool_op_get_tso(args);
	else
		netdev->ethtool_ops->get_tso(args);

Because this certainly seems like desirable behavior across all devices.

Re: [PATCH 1/6] ibmveth: Enable TCP checksum offload

[Jeff Garzik]

Brian King wrote:
> This patchset enables TCP checksum offload support for IPV4
> on ibmveth. This completely eliminates the generation and checking of
> the checksum for packets that are completely virtual and never
> touch a physical network. A simple TCP_STREAM netperf run on
> a virtual network with maximum mtu set yielded a ~30% increase
> in throughput. This feature is enabled by default on systems that
> support it, but can be disabled with a module option.
> 
> Signed-off-by: Brian King <brking@linux.vnet.ibm.com>

applied 1-6