Re: Is this a bug in TCP or the PCNet32 driver?

[Andrew Morton]

Dave Larson wrote:
> 
> The way things are today, the TCP code rely on the hardware drivers to
> free an skb as soon as it is transmitted. But in that case of PCNet32,
> that doesn't happen. On the other hand, PCNet32 does seem reasonable in
> it's attempts to reduce the number of interrupts, although that breaks the
> tcp code in this case were these isn't much network activity.
> 

Yup.  Tx interrupt mitigation like this is a really neat feature. It
can make a huge improvement in performance.  But the driver does need
to implement a timer to fix the problem which you have described.

-