From: Andreas Oberritter <[email protected]>
The write barrier should be used before starting a DMA transfer. This fixes
a problem, where almost all packets received on another machine had garbled
content. Tested with an RTL8100C on a MIPS machine.
Signed-off-by: Andreas Oberritter <[email protected]>
diff --git a/drivers/net/8139too.c b/drivers/net/8139too.c
index 53bd903..44344af 100644
--- a/drivers/net/8139too.c
+++ b/drivers/net/8139too.c
@@ -1716,13 +1716,13 @@ static int rtl8139_start_xmit (struct sk_buff *skb, struct net_device *dev)
}
spin_lock_irqsave(&tp->lock, flags);
+ wmb();
RTL_W32_F (TxStatus0 + (entry * sizeof (u32)),
tp->tx_flag | max(len, (unsigned int)ETH_ZLEN));
dev->trans_start = jiffies;
tp->cur_tx++;
- wmb();
if ((tp->cur_tx - NUM_TX_DESC) == tp->dirty_tx)
netif_stop_queue (dev);
Andreas Oberritter wrote:
> From: Andreas Oberritter <[email protected]>
>
> The write barrier should be used before starting a DMA transfer. This fixes
> a problem, where almost all packets received on another machine had garbled
> content. Tested with an RTL8100C on a MIPS machine.
>
> Signed-off-by: Andreas Oberritter <[email protected]>
>
> diff --git a/drivers/net/8139too.c b/drivers/net/8139too.c
> index 53bd903..44344af 100644
> --- a/drivers/net/8139too.c
> +++ b/drivers/net/8139too.c
> @@ -1716,13 +1716,13 @@ static int rtl8139_start_xmit (struct sk_buff *skb, struct net_device *dev)
> }
>
> spin_lock_irqsave(&tp->lock, flags);
> + wmb();
[...]
All memory barriers need a comment to explain why and what they're doing.
Ben.
--
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
From: Andreas Oberritter <[email protected]>
The write barrier should be used before starting a DMA transfer. This fixes
a problem, where almost all packets received on another machine had garbled
content. Tested with an RTL8100C on a MIPS machine.
Signed-off-by: Andreas Oberritter <[email protected]>
---
Added a comment on Ben's request explaining the reason for the barrier.
diff --git a/drivers/net/8139too.c b/drivers/net/8139too.c
index 53bd903..3c5fac6 100644
--- a/drivers/net/8139too.c
+++ b/drivers/net/8139too.c
@@ -1716,13 +1716,18 @@ static int rtl8139_start_xmit (struct sk_buff *skb, struct net_device *dev)
}
spin_lock_irqsave(&tp->lock, flags);
+ /*
+ * Writing to TxStatus triggers a DMA transfer of the data
+ * copied to tp->tx_buf[entry] above. Use a memory barrier
+ * to make sure that the device sees the updated data.
+ */
+ wmb();
RTL_W32_F (TxStatus0 + (entry * sizeof (u32)),
tp->tx_flag | max(len, (unsigned int)ETH_ZLEN));
dev->trans_start = jiffies;
tp->cur_tx++;
- wmb();
if ((tp->cur_tx - NUM_TX_DESC) == tp->dirty_tx)
netif_stop_queue (dev);