hyperv/hn: Partly rework transmission path

[sepherosa_gmail.com]

Summary

• Avoid unnecessary malloc/free on transmission path.
• busdma(9)-fy transmission path.
• Properly handle IFF_DRV_OACTIVE.
• Properly setup TSO parameters.
• Properly handle bpf(4) tapping.
• Allow size of chimney sending be tuned on a running system. Default value still needs more test to determine.

[sepherosa_gmail.com]

If no objection comes, this will be committed two days later.

[delphij]

sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c

729

Do we need to assert the return value here?

982

Please consider comparing with 0 explicitly.

1034

The return value is not asserted here -- is it intentional?

1066

Suggestion: compare 'freed' with 0 explicitly.

[sepherosa_gmail.com]

sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c

729

Nope, we should not assert it here; since it may not be the last txdesc_put, the txdesc_put in hn_start_locked() (for the BPF tap) may be the last txdesc_put.

982

ok

1034

Intentional, same reason as the txdesc_put in netvsc_xmit_completion().

1066

sure :)

[sepherosa_gmail.com]

Test freed against 0. Suggested by delphij

[adrian]

So I'm okay with this at a first glance, but next time please break it up into smaller pieces. Stuff like busdma'ing things and adding counters should really be separate commits. :-)

So, +1 from me, I'm assuming you've tested it, let's use smaller commits next time!

[sepherosa_gmail.com]

In D4972#106738, @adrian wrote:

So I'm okay with this at a first glance, but next time please break it up into smaller pieces. Stuff like busdma'ing things and adding counters should really be separate commits. :-)

So, +1 from me, I'm assuming you've tested it, let's use smaller commits next time!

OK, I will try next time. It's a big patch, but the original TX path is just too broken to have this patch splitted ;)