In article <200504181001.13399.scw@netbsd.org>
scw@netbsd.org wrote:

> An easy way to provoke this bug is to drop into ddb, then ping the 
> machine until the fxp's Rx buffers are exhausted.

I wonder if the last RFA should have FXP_RFA_CONTROL_S in
rfa_contrl to suspend Rx ops when RFA is exhausted.

I also wonder if we should take these OpenBSD's fixes:
http://www.openbsd.org/cgi-bin/cvsweb/src/sys/dev/ic/fxp.c#rev1.61
http://www.openbsd.org/cgi-bin/cvsweb/src/sys/dev/ic/fxpreg.h#rev1.7
---
Izumi Tsutsui