Update to -STABLE breaks re(4)

[Scott Long]

Martin wrote:
> Hi,
> 
> today I've compiled new world and kernel. Last kernel was
> from November, I think.
> 
> The platform is amd64, mainboard Gigabyte K8VNXP. I'm
> talking about -STABLE here (RELENG_5 tag).
> 
> (I've deleted /usr/obj before making world and used
> the usual procedure to update.)
> 
> After reboot, re(4) "Realtek RTL8110S" could not be detected
> automatically. It is statically compiled into the kernel.
> The dmesg said something like "Could not send TCP packet in
> loopback mode" or similar.
> 
> I tried "kldload if_re". It worked fine (e.a. the interface
> works after /etc/netstart restart), but I got few
> warnings ("already exists").
> 
> Next, I added if_re_load="YES" to /boot/loader.conf (tried
> a workaround). The PC always rebooted before reaching the
> "beastie menu".
> 
> I can boot now using /boot/loader.old and the
> re(4)-module seems to load with same warnings as above.
> 
> If I remove if_re_load="YES", even using the old loader
> my re0 won't be detected (same problem as above).
> 
> Conclusion:
> It seems there are two problems. The new loader seems
> to crash while loading if_re (maybe because of the handling
> of the warning). And something changed in the detection routine
> of the re(4) driver, which broke support for RTL8110S.
> 
> --
> 
> I won't have access to this PC until after New Year, so it
> will be difficult to do further research. I can imagine that
> there are more people who have this mainboard, so maybe 
> someone can confirm the problem?
> 
> Martin
> 

This might be another victim of my busdma fixes.  There are definite 
mistakes in the if_re driver that I need to fix; let me know when you
have some time to test them and I'll generate patches.

Scott