fix(fwe): add missing net epoch around ether_input

[guest-seuros]

Summary

Wrap the if_input() call in fwe_as_input() with NET_EPOCH_ENTER/EXIT.
The network stack requires epoch protection when delivering packets
via if_input, and fwe was missing it.

[adrian]

hm, good question, should we do it per frame or in batches?

[zlei]

In D57459#1316924, @adrian wrote:

hm, good question, should we do it per frame or in batches?

A bit late to see this patch.

The NET_EPOCH_ENTER() / NET_EPOCH_EXIT() is a bit heavy weight on every packet. For high speed hardware you want to batch input the packets. For firewire I think it is not a high speed hardware so that ( batch input ) probably make no noticeable difference.

[adrian]

In D57459#1317325, @zlei wrote:

In D57459#1316924, @adrian wrote:

hm, good question, should we do it per frame or in batches?

A bit late to see this patch.

The NET_EPOCH_ENTER() / NET_EPOCH_EXIT() is a bit heavy weight on every packet. For high speed hardware you want to batch input the packets. For firewire I think it is not a high speed hardware so that ( batch input ) probably make no noticeable difference.

Yeah, I've asked seuros about reproducing it, we can clean this up in a follow-up commit (populate a work list in the loop, then do enter/loop-if-input/exit).

I wanted the code in -head to work first, then we can make it faster-er. :-)

Thanks for reviewing it tho!

[glebius]

Out of curiosity: what hardware do you use to encounter this bug? Is it legacy (about 15 year old) mainboard or is it something new like https://www.startech.com/en-us/cards-adapters/firewire?

[guest-seuros]

Apple Computers (Imacs, MacMini, Laptops), Lenovo T61p, Startech, SIIG, LACIE.

I daisy chained 3 computers and tried to push a tar file from computer 1 to 3 with a fw800 or fw400 link.

The panic happens with any revision.