David Miller [off-list ref] writes:

From: ebiederm@xmission.com (Eric W. Biederman)
Date: Tue, 11 Mar 2014 01:43:10 -0700

quoted

Furthermore netpoll by it's design depends on the ability to receive
packets in netpoll_poll_dev.  It is a capability I don't think we have
ever used in the mainline kernel but it is a capability that is there
deliberately.  Which means if we want netpoll to not mess with the rx
path we need to change netpoll.

This breaks kdump, and any other users of netpoll_rx() et al.

It does not break kdump.  (kdump starts a new kernel to do it's work). 

It does break the ancient lkcd netdump that was never merged, and has
been abandoned (to the best of my knowledge). crash dumps proved
entirely too fragile to perform from a broken kernel.

It does break kgdboe that was never merged.

Make the zero budget depend upon us being invoked from hardware
irq context, or something like that.

Good enough.  I will respin my driver patches based on the assumption
that netpoll will be changed in this way.  There are no dependencies
for the drivers, I just need to remove my rx path changes.

We can have the conversation about how to change netpoll in parallel.

Eric