Committed as 954712e8f6.

@erj was the original change to this intended? Before the change it was:

Address feedback from @zlei and @np.

Replaced by D38735.

Adding @jmg, since he made the change to ure(4) to change the drv queue length to 512.

Thanks!

Code was rewritten.

In D38714#881069, @dchagin wrote:

@jhibbits, I also reverted 52435989, as linprocfs code also needed to update

In D38609#880936, @dchagin wrote:

With two last patches the kernel paniced: Feb 21 19:27:01 mordor savecore[1488]: reboot after panic: Assertion in_epoch(net_epoch_preempt) failed at /home/dchagin/freebsd/sys/net/if.c:4520

As we found some problem in Linuxulator netlink which is merged to stable/13 branch and requires simple change to the code modified by a 52d98483 commit, I propose to revert 52d98483 until it addressed, and on a weekend I’ll help to fix it.

Rebase, now based on fix in D38703, too.

In D38609#880820, @dchagin wrote:

latest diff is not cleanly applies:

Change summary to 'linux' instead of 'IfAPI'

Address feedback from @dchagin and @melifaro

In D38609#878650, @glebius wrote:

Wait a second... Are we sure that IFNET_RLOCK() isn't enough for that access? Maybe we need to adjust the assertion that we recently added? That it asserts epoch OR the lock. @melifaro What do you think?

Not coccinelle, but there's a script that'll get "most of the way" at tools/ifnet/convert_ifapi.sh . Coccinelle would probably be better, less error-prone, but I don't know enough, and the script has been in the tree since 2014 (as convert_drvapi.sh)

In D38609#878624, @glebius wrote:

The revision header needs to be changed s/IfAPI:/linux:/

Address @zlei's feedback.

Address @zlei's feedback.

One more whitespace bug.

Address more feedback from @kib.

In D38499#877645, @glebius wrote:

Didn't mean to override your comments! I'm pretty sure @jhibbits will address them before checking in.

Address @kib's feedback. I think I got them all.

Address feedback from @kib.

In D38499#876078, @kib wrote:

A lot of functions return int and that boils down to return (0);. Do you intend to make the fail in some way in the future? If yes, could you please describe what are the expected failure modes?

Address feedback. Clean up the pointless casts.

Address feedback. No more DRVAPI comments in the code.

I don't know how any of the issues you pointed out compiled in the first place. No errors on them. Will post an updated diff shortly.

I like the idea of doing less work in interrupt handlers. Especially when it could potentially share a cache line.