hme(4) broken on non-sparc64 systems in -current

[Pyun YongHyeon]

On Mon, Mar 20, 2006 at 10:17:21PM +0100, Joerg Wunsch wrote:
 > Even though this problem is not related to sparc64 system (I hope, the
 > least), I think all those who know about hme(4) are listening here.
 > 
 > I used to run a PCI QFE card in my (i386) scratch machine at home, for
 > various experiments.  After upgrading the machine from a 6-stable of
 > about a year ago to -current, the QFE ceased to work.  The symptoms
 > were that I could still ping anyone (even with large packets), but all
 > TCP and UDP traffic seemingly didn't ``arrive'' at the IP stack, even
 > though the packets made it into the NIC at the lowest level (so
 > tcpdump could still display them).
 > 
 > This made me suspect the TCP/UDP checksum offloading, and indeed,
 > after uncommenting the checksum capability:
 > 
 > Index: if_hme.c
 > ===================================================================
 > RCS file: /home/ncvs/src/sys/dev/hme/if_hme.c,v
 > retrieving revision 1.46
 > diff -u -r1.46 if_hme.c
 > --- if_hme.c    17 Jan 2006 06:02:22 -0000      1.46
 > +++ if_hme.c    20 Mar 2006 20:56:54 -0000
 > @@ -340,9 +340,9 @@
 >          * Tell the upper layer(s) we support long frames/checksum offloads.
 >          */
 >         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
 > -       ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
 > +       ifp->if_capabilities |= IFCAP_VLAN_MTU /* | IFCAP_HWCSUM */;
 >         ifp->if_hwassist |= sc->sc_csum_features;
 > -       ifp->if_capenable |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
 > +       ifp->if_capenable |= IFCAP_VLAN_MTU /* | IFCAP_HWCSUM */;
 >         return (0);
 >  
 >  fail_txdesc:
 > 
 > everything works again.  If anyone has any further ideas what might
 > have broken this, I'm all ears, otherwise I might start digging down
 > into the code myself.
 > 
 > (I don't have a FreeBSD-sparc64 machine running -current around, so I
 > cannot test right now whether it would work there.)
 > 

How about backing out rev. 1.46(if_hme.c)?
For the same patch sent to bard at OpenBSD I got a positive report
so it's strange to me though.(brad at OpenBSD reported Rx checksum
offload breakage on little endian systems.)
Since I don't have PCI HME NIC I can't test it on little endian
systems.  :-(

-- 
Regards,
Pyun YongHyeon