Do others think it would be helpful for dev->hard_start_xmit() to know
if the packet being Tx'd is the last in a sequence?
I was thinking that the net driver might be able to make some
optimizations if it knew that other packet were (or were not) to follow,
after the current one being processed.
In general processing items in a series usually lends to a certain
amount of optimization. When Tulip chips are using the ring (not linked
list) style of descriptors, you have two not one buffers for each
descriptor. Processing things in a series also may allow one to make
better use of various interrupt mitigation strategies.
Jeff
--
Jeff Garzik |
Building 1024 | Yossarian lives.
MandrakeSoft, Inc. |
|