David Edmondson wrote:
On 2 Jun 2006, at 9:12pm, Adi Masputra wrote:
In any case, the stack is more flexible for inbound packets.
As long as you negotiate *any* kind of hardware checksum
offload, then the full/partial/none checksum info is on
a per-packet basis.
Though, in itself this can be a pain. If I have an interface that can
offload receive side checksum calculation but not transmit (perhaps the
transmit hardware is broken) I'm stuck, as the receive side offload is
ignored unless I claim to be able to do transmit side offload.
This (requiring the ability to perform transmit side offload to permit
receive side offload) is new with Yosemite, as it never used to be the
case. What was the rationale?
I don't recall that such a behavior was added in Yosemite,
although I have no way of confirming this now. There's a
check in the stack to see if the packet comes from an
interface that has negotiated the offload feature during
plumbing time; this isn't specific to hardware checksum
and I believe it has been there pre-Yosemite.
In reality, this situation is a corner case one, and NICs
typically share the same checksum engine/logic for both
directions -- if one direction has a bug then the driver
may as well disable both.
Adi
_______________________________________________
networking-discuss mailing list
[email protected]
