Hello Sverre. While doing your testing, also watch the output of
netstat -s. Specifically, We're interested to see if you see a large
number of packets dropped due to checksum failures. The patch I provided
affects tcp and udp packets. Look for output similar to the following:
netstat -s
...
[ ... ]
tcp:
[ ... ]
6233676 window update packets
73755 packets received after close
36 discarded for bad checksums
0 discarded for bad header offset fields
[ ... ]
udp:
6513121 datagrams received
0 with incomplete header
0 with bad data length field
0 with bad checksum
538021 dropped due to no socket
181641 broadcast/multicast datagrams dropped due to no socket
0 dropped due to full socket buffers
5793459 delivered
539186 PCB hash misses
6158353 datagrams output
On Sep 25, 10:56am, Sverre Froyen wrote:
} Subject: Re: Fixes for kern/40018 -- any chance of getting these pulled in
} On Tue September 21 2010 18:09:08 Brian Buhrow wrote:
} > Hello. Excellent! Let me know if you have trouble getting the
} > patches from the bug report and/or getting them into a kernel.
}
} With the patch applied to NetBSD 5.1_RC4 (cvs update -r netbsd-5) and
} checksums turned on:
}
} # dmesg | grep bge0
} bge0 at pci1 dev 6 function 0: Broadcom BCM5701 Gigabit Ethernet
} bge0: interrupting at ioapic1 pin 12
} bge0: ASIC BCM5701 B5 (0x0105), Ethernet address 00:06:5b:fd:79:a6
} brgphy0 at bge0 phy 1: BCM5701 1000BASE-T media interface, rev. 0
}
} # ifconfig bge0
} bge0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
}
capabilities=3f00<IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx>
}
enabled=3f00<IP4CSUM_Rx,IP4CSUM_Tx,TCP4CSUM_Rx,TCP4CSUM_Tx,UDP4CSUM_Rx,UDP4CSUM_Tx>
} address: 00:06:5b:fd:79:a6
} media: Ethernet autoselect (100baseTX full-
} duplex,flowcontrol,rxpause,txpause)
} status: active
} inet xxx.xxx.xxx.xxx netmask 0xffffff00 broadcast xxx.xxx.xxx.255
} inet6 xxxx::xxx:xxxx:xxxx:xxxx%bge0 prefixlen 64 scopeid 0x1
}
} the server has been running without any problems for several days. Tests have
} included repeated copies of large files (~200 MB) to the server using scp and
} NFS (with the server as a client). The server has also survived full backups.
} Transfer speeds are at wire speed (11 MB/s for scp, 12 MB/s for NFS).
}
} Please let me know if I should test anything else (unfortunately, I'm unable
} to test Gb/s speed).
}
} Regards,
} Sverre
}
}
}
>-- End of excerpt from Sverre Froyen
