On Wed, 31 May 2006, Simon Leinen wrote:
I recently upgraded my SunBlade 2500 from Nevada build33 to build38, and found that I could no longer communicate with other hosts over TCP over IPv6. The SunBlade sends out packets, but never receives any responses. So I logged in to another machine and captured packets while attempting a connection from the SunBlade. The SYN packets are seen on the other machine, but according to Ethereal, they have incorrect TCP checksums. For the (small) packets I observed in this way, the incorrect checksums always seem to be of the form 0xDFxy (xy varies). I could observe the same issue with UDP-over-IPv6 checksums. TCP-over-IPv4 and UDP-over-IPv4 work fine, and ICMPv6 checksums are also correct.
Think this is a known issue on some bge chipsets. (See Bug id 6400907 via http://bugs.opensolaris.org/bugdatabase/index.jsp)
I assume that the new snapshot is trying to offload checksum computation for TCP/UDP over IPv6 to the Broadcom (bge) interface. That would sound like a new feature from the "Yosemite" project. Unfortunately this doesn't seem to work well in all cases. Can I somehow suppress checksum offload in order to test this hypothesis?
add "set ip:dohwcksum=0" in /etc/system -venu
-- Simon. This message posted from opensolaris.org _______________________________________________ networking-discuss mailing list [email protected]
_______________________________________________ networking-discuss mailing list [email protected]
