Skip to contentSkip to conversation

ibmveth: Fix more little endian issues

PATCH In review Alexander Graf 6 participants 9 messages Dec 23, 2013 – Mar 6, 2014

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ibmveth: Fix more little endian issues

The ibmveth driver is memcpy()'ing the mac address between a variable (register) and memory. This assumes a certain endianness of the system, so let's make that implicit assumption work again.

This patch adds be64_to_cpu() calls to all places where the mac address gets memcpy()'ed into a local variable.

---
 drivers/net/ethernet/ibm/ibmveth.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/ibm/ibmveth.c +2 -1
index 952d795..97f4ee96 100644
@@ -581,7 +581,7 @@ static int ibmveth_open(struct net_device *netdev)
 	adapter->rx_queue.toggle = 1;
 
 	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
-	mac_address = mac_address >> 16;
+	mac_address = be64_to_cpu(mac_address) >> 16;
 
 	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 					adapter->rx_queue.queue_len;
@@ -1186,6 +1186,7 @@ static void ibmveth_set_multicast_list(struct net_device *netdev)
 			/* add the multicast address to the filter table */
 			unsigned long mcast_addr = 0;
 			memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
+			mcast_addr = cpu_to_be64(mcast_addr);
 			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
 						   IbmVethMcastAddFilter,
 						   mcast_addr);
Signature · 1 line
1.8.1.4
#1 lore raw

Hi Alex,

Quoted 3 lines

The ibmveth driver is memcpy()'ing the mac address between a variable (register) and memory. This assumes a certain endianness of the system, so let's make that implicit assumption work again.

Nice catch! I don't like how the driver has two different methods for creating these MAC addresses, both without comments. How does this look?

Anton
--

The hypervisor expects MAC addresses passed in registers to be big endian u64. Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

---

drivers/net/ethernet/ibm/ibmveth.c +15 -8
index 952d795..044178b 100644
@@ -523,6 +523,17 @@ retry:
 	return rc;
 }
 
+/*
+ * The hypervisor expects MAC addresses passed in registers to be
+ * big endian u64.
+ */
+static unsigned long ibmveth_encode_mac_addr(char *mac)
+{
+	unsigned long encoded = 0;
+	memcpy(((char *)&encoded) + 2, mac, ETH_ALEN);
+	return cpu_to_be64(encoded);
+}
+
 static int ibmveth_open(struct net_device *netdev)
 {
 	struct ibmveth_adapter *adapter = netdev_priv(netdev);
@@ -580,8 +591,7 @@ static int ibmveth_open(struct net_device *netdev)
 	adapter->rx_queue.num_slots = rxq_entries;
 	adapter->rx_queue.toggle = 1;
 
-	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
-	mac_address = mac_address >> 16;
+	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 
 	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 					adapter->rx_queue.queue_len;
@@ -1184,8 +1194,8 @@ static void ibmveth_set_multicast_list(struct net_device *netdev)
 		/* add the addresses to the filter table */
 		netdev_for_each_mc_addr(ha, netdev) {
 			/* add the multicast address to the filter table */
-			unsigned long mcast_addr = 0;
-			memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
+			unsigned long mcast_addr;
+			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
 			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
 						   IbmVethMcastAddFilter,
 						   mcast_addr);
@@ -1369,9 +1379,6 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 
 	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
 
-	adapter->mac_addr = 0;
-	memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);
-
 	netdev->irq = dev->irq;
 	netdev->netdev_ops = &ibmveth_netdev_ops;
 	netdev->ethtool_ops = &netdev_ethtool_ops;
@@ -1380,7 +1387,7 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 	netdev->features |= netdev->hw_features;
 
-	memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
+	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
 
 	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
drivers/net/ethernet/ibm/ibmveth.h +0 -1
index 84066ba..2c636cb 100644
@@ -139,7 +139,6 @@ struct ibmveth_adapter {
     struct napi_struct napi;
     struct net_device_stats stats;
     unsigned int mcastFilterSize;
-    unsigned long mac_addr;
     void * buffer_list_addr;
     void * filter_list_addr;
     dma_addr_t buffer_list_dma;
author Reviewed-by
#2 lore raw
Quoted 10 lines · Anton Blanchard

Hi Alex,

Quoted 3 lines

The ibmveth driver is memcpy()'ing the mac address between a variable (register) and memory. This assumes a certain endianness of the system, so let's make that implicit assumption work again.

Nice catch! I don't like how the driver has two different methods for creating these MAC addresses, both without comments. How does this look?

Heh - I didn't even realize those two places were doing the same thing.

Obviously your patch is by far nicer.

Alex

Quoted 90 lines

Anton
--

The hypervisor expects MAC addresses passed in registers to be big endian u64. Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

---

drivers/net/ethernet/ibm/ibmveth.c +15 -8
index 952d795..044178b 100644
@@ -523,6 +523,17 @@ retry:
 	return rc;
 }
 
+/*
+ * The hypervisor expects MAC addresses passed in registers to be
+ * big endian u64.
+ */
+static unsigned long ibmveth_encode_mac_addr(char *mac)
+{
+	unsigned long encoded = 0;
+	memcpy(((char *)&encoded) + 2, mac, ETH_ALEN);
+	return cpu_to_be64(encoded);
+}
+
 static int ibmveth_open(struct net_device *netdev)
 {
 	struct ibmveth_adapter *adapter = netdev_priv(netdev);
@@ -580,8 +591,7 @@ static int ibmveth_open(struct net_device *netdev)
 	adapter->rx_queue.num_slots = rxq_entries;
 	adapter->rx_queue.toggle = 1;
 
-	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
-	mac_address = mac_address >> 16;
+	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 
 	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 					adapter->rx_queue.queue_len;
@@ -1184,8 +1194,8 @@ static void ibmveth_set_multicast_list(struct net_device *netdev)
 		/* add the addresses to the filter table */
 		netdev_for_each_mc_addr(ha, netdev) {
 			/* add the multicast address to the filter table */
-			unsigned long mcast_addr = 0;
-			memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
+			unsigned long mcast_addr;
+			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
 			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
 						   IbmVethMcastAddFilter,
 						   mcast_addr);
@@ -1369,9 +1379,6 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 
 	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
 
-	adapter->mac_addr = 0;
-	memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);
-
 	netdev->irq = dev->irq;
 	netdev->netdev_ops = &ibmveth_netdev_ops;
 	netdev->ethtool_ops = &netdev_ethtool_ops;
@@ -1380,7 +1387,7 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 	netdev->features |= netdev->hw_features;
 
-	memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
+	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
 
 	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
drivers/net/ethernet/ibm/ibmveth.h +0 -1
index 84066ba..2c636cb 100644
@@ -139,7 +139,6 @@ struct ibmveth_adapter {
    struct napi_struct napi;
    struct net_device_stats stats;
    unsigned int mcastFilterSize;
-    unsigned long mac_addr;
    void * buffer_list_addr;
    void * filter_list_addr;
    dma_addr_t buffer_list_dma;
#3 lore raw
In reply to Anton Blanchard
Quoted 2 lines · Anton Blanchard

The hypervisor expects MAC addresses passed in registers to be big endian u64.

So maybe use __be64 declarations?

Quoted 1 line

+static unsigned long ibmveth_encode_mac_addr(char *mac)

static __be64 ibmveth_encode_mac_addr(const char *mac)

?

etc...

#4 lore raw

PATCH 1/1 ibmveth: Fix more little endian issues

The hypervisor expects MAC addresses passed in registers to be big endian u64. Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

---

v2: annotate with __be64 as suggested by Joe

drivers/net/ethernet/ibm/ibmveth.c +16 -10
index 952d795..bb9a631 100644
@@ -497,7 +497,7 @@ static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 }
 
 static int ibmveth_register_logical_lan(struct ibmveth_adapter *adapter,
-        union ibmveth_buf_desc rxq_desc, u64 mac_address)
+			union ibmveth_buf_desc rxq_desc, __be64 mac_address)
 {
 	int rc, try_again = 1;
 
@@ -523,10 +523,20 @@ retry:
 	return rc;
 }
 
+/* The hypervisor expects MAC addresses passed in registers to be
+ * big endian u64.
+ */
+static __be64 ibmveth_encode_mac_addr(char *mac)
+{
+	unsigned long encoded = 0;
+	memcpy(((char *)&encoded) + 2, mac, ETH_ALEN);
+	return cpu_to_be64(encoded);
+}
+
 static int ibmveth_open(struct net_device *netdev)
 {
 	struct ibmveth_adapter *adapter = netdev_priv(netdev);
-	u64 mac_address = 0;
+	__be64 mac_address = 0;
 	int rxq_entries = 1;
 	unsigned long lpar_rc;
 	int rc;
@@ -580,8 +590,7 @@ static int ibmveth_open(struct net_device *netdev)
 	adapter->rx_queue.num_slots = rxq_entries;
 	adapter->rx_queue.toggle = 1;
 
-	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
-	mac_address = mac_address >> 16;
+	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 
 	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 					adapter->rx_queue.queue_len;
@@ -1184,8 +1193,8 @@ static void ibmveth_set_multicast_list(struct net_device *netdev)
 		/* add the addresses to the filter table */
 		netdev_for_each_mc_addr(ha, netdev) {
 			/* add the multicast address to the filter table */
-			unsigned long mcast_addr = 0;
-			memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
+			__be64 mcast_addr;
+			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
 			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
 						   IbmVethMcastAddFilter,
 						   mcast_addr);
@@ -1369,9 +1378,6 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 
 	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
 
-	adapter->mac_addr = 0;
-	memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);
-
 	netdev->irq = dev->irq;
 	netdev->netdev_ops = &ibmveth_netdev_ops;
 	netdev->ethtool_ops = &netdev_ethtool_ops;
@@ -1380,7 +1386,7 @@ static int ibmveth_probe(struct vio_dev *dev, const struct vio_device_id *id)
 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 	netdev->features |= netdev->hw_features;
 
-	memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
+	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
 
 	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
drivers/net/ethernet/ibm/ibmveth.h +0 -1
index 84066ba..2c636cb 100644
@@ -139,7 +139,6 @@ struct ibmveth_adapter {
     struct napi_struct napi;
     struct net_device_stats stats;
     unsigned int mcastFilterSize;
-    unsigned long mac_addr;
     void * buffer_list_addr;
     void * filter_list_addr;
     dma_addr_t buffer_list_dma;
#5 lore raw
Quoted 7 lines · Anton Blanchard

The hypervisor expects MAC addresses passed in registers to be big endian u64. Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

[...]

drivers/net/ethernet/ibm/ibmveth.c retry: · line 531

Quoted 10 lines
drivers/net/ethernet/ibm/ibmveth.c +6 -0
@@ -523,10 +523,20 @@ retry:
 	return rc;
 }
 
+/* The hypervisor expects MAC addresses passed in registers to be
+ * big endian u64.
+ */
+static __be64 ibmveth_encode_mac_addr(char *mac)
+{
+	unsigned long encoded = 0;

u64

drivers/net/ethernet/ibm/ibmveth.c retry: · line 534

Quoted 3 lines
drivers/net/ethernet/ibm/ibmveth.c +3 -0
+	memcpy(((char *)&encoded) + 2, mac, ETH_ALEN);
+	return cpu_to_be64(encoded);
+}

[...]

So on big-endian systems the byte order of the result will be:

0 0 mac0 mac1 mac2 mac3 mac4 mac5

and on little-endian systems it's:

mac5 mac4 mac3 mac2 mac1 mac0 0 0

It seems to me that 'encoded' is actually in big-endian order and this function returns the address in CPU order.

So are you sure your explanation isn't backwards, because it looks to me like the driver was already holding the MAC address in big-endian order and perhaps the hypercall mechanism does a byte-swap when the guest is little-endian.

Ben.

Signature · 3 lines
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
#6 lore raw

ibmveth: Fix endian issues with MAC addresses

In reply to Anton Blanchard

The code to load a MAC address into a u64 for passing to the hypervisor via a register is broken on little endian.

Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

---

drivers/net/ethernet/ibm/ibmveth.c +16 -9
Index: b/drivers/net/ethernet/ibm/ibmveth.c
===================================================================
@@ -522,10 +522,21 @@ retry:
 	return rc;
 }
 
+static u64 ibmveth_encode_mac_addr(u8 *mac)
+{
+	int i;
+	u64 encoded = 0;
+
+	for (i = 0; i < ETH_ALEN; i++)
+		encoded = (encoded << 8) | mac[i];
+
+	return encoded;
+}
+
 static int ibmveth_open(struct net_device *netdev)
 {
 	struct ibmveth_adapter *adapter = netdev_priv(netdev);
-	u64 mac_address = 0;
+	u64 mac_address;
 	int rxq_entries = 1;
 	unsigned long lpar_rc;
 	int rc;
@@ -579,8 +590,7 @@ static int ibmveth_open(struct net_devic
 	adapter->rx_queue.num_slots = rxq_entries;
 	adapter->rx_queue.toggle = 1;
 
-	memcpy(&mac_address, netdev->dev_addr, netdev->addr_len);
-	mac_address = mac_address >> 16;
+	mac_address = ibmveth_encode_mac_addr(netdev->dev_addr);
 
 	rxq_desc.fields.flags_len = IBMVETH_BUF_VALID |
 					adapter->rx_queue.queue_len;
@@ -1183,8 +1193,8 @@ static void ibmveth_set_multicast_list(s
 		/* add the addresses to the filter table */
 		netdev_for_each_mc_addr(ha, netdev) {
 			/* add the multicast address to the filter table */
-			unsigned long mcast_addr = 0;
-			memcpy(((char *)&mcast_addr)+2, ha->addr, ETH_ALEN);
+			u64 mcast_addr;
+			mcast_addr = ibmveth_encode_mac_addr(ha->addr);
 			lpar_rc = h_multicast_ctrl(adapter->vdev->unit_address,
 						   IbmVethMcastAddFilter,
 						   mcast_addr);
@@ -1372,9 +1382,6 @@ static int ibmveth_probe(struct vio_dev
 
 	netif_napi_add(netdev, &adapter->napi, ibmveth_poll, 16);
 
-	adapter->mac_addr = 0;
-	memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);
-
 	netdev->irq = dev->irq;
 	netdev->netdev_ops = &ibmveth_netdev_ops;
 	netdev->ethtool_ops = &netdev_ethtool_ops;
@@ -1383,7 +1390,7 @@ static int ibmveth_probe(struct vio_dev
 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 	netdev->features |= netdev->hw_features;
 
-	memcpy(netdev->dev_addr, &adapter->mac_addr, netdev->addr_len);
+	memcpy(netdev->dev_addr, mac_addr_p, ETH_ALEN);
 
 	for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
 		struct kobject *kobj = &adapter->rx_buff_pool[i].kobj;
drivers/net/ethernet/ibm/ibmveth.h +0 -1
Index: b/drivers/net/ethernet/ibm/ibmveth.h
===================================================================
@@ -138,7 +138,6 @@ struct ibmveth_adapter {
     struct napi_struct napi;
     struct net_device_stats stats;
     unsigned int mcastFilterSize;
-    unsigned long mac_addr;
     void * buffer_list_addr;
     void * filter_list_addr;
     dma_addr_t buffer_list_dma;
maintainer Applied
#7 lore raw

Re: ibmveth: Fix endian issues with MAC addresses

From: Anton Blanchard <[email protected]>
Date: Wed, 5 Mar 2014 14:51:37 +1100

Quoted 6 lines

The code to load a MAC address into a u64 for passing to the hypervisor via a register is broken on little endian.

Create a helper function called ibmveth_encode_mac_addr which does the right thing in both big and little endian.

Quoted 6 lines

We were storing the MAC address in a long in struct ibmveth_adapter. It's never used so remove it - we don't need another place in the driver where we create endian issues with MAC addresses.

Applied, thanks Anton.

Quoted 1 line
drivers/net/ethernet/ibm/ibmveth.c +0 -1
-	memcpy(&adapter->mac_addr, mac_addr_p, ETH_ALEN);

...

Quoted 1 line

- unsigned long mac_addr;

That's some scary stuff right there.

#8 lore raw
In reply to Joe Perches
Quoted 9 lines · Joe Perches
Quoted 2 lines · Anton Blanchard

The hypervisor expects MAC addresses passed in registers to be big endian u64.

So maybe use __be64 declarations?

Quoted 1 line

+static unsigned long ibmveth_encode_mac_addr(char *mac)

static __be64 ibmveth_encode_mac_addr(const char *mac)

A register value has no endianness. Only memory content does. Especially talking of a MAC address which is really a byte stream.... (Yes, our __beXX types used without a * are borderline, but we've got used to it).

In fact I find the use of

    memcpy(((char *)&encoded) + 2, mac, ETH_ALEN);

Really gross :-) Yes it works with the added cpu_to_be64() but in that specific case, I think it would be nicer to simply load & shift into position the 6 bytes and avoid the endianness issue completely.

Cheers,
Ben.