Re: Packet corruption in re0

On Thu, Feb 21, 2008 at 07:43:18AM +0000, Niki Denev wrote:
 > On Thu, Feb 21, 2008 at 5:15 AM, Eric L. Chen <d9364104_at_mail.nchu.edu.tw> wrote:
 > >
 > >
 > >  On Thu, 2008-02-21 at 11:03 +1000, Robert Backhaus wrote:
 > >  > I am experiencing roughly 15% packet corruption on the re interface on
 > >  > my freebsd 7/amd64  box.
 > >  >
 > >  > FreeBSD gw.flexi.robbak.com 7.0-PRERELEASE FreeBSD 7.0-PRERELEASE #8:
 > >  > Tue Feb  5 09:49:55 EST 2008
 > >  > root_at_gw.flexi.robbak.com:/usr/obj/usr/src/sys/GW  amd64
 > >  >
 > >  > The attached 3 files demonstrate the problem: "ping" shows the output
 > >  > of ping -c 100, and shows 15% packet loss. "tcpdump" shows the packets
 > >  > leaving, and some of the lost packets being returned with addresses,
 > >  > ports and data corrupted. The data in these packets seems to be coming
 > >  > from other packets passing through other interfaces at the time.
 > >  > "remote-tcpdump" shows the packets being received and returned from
 > >  > the other machine. Note that some packets are being corrupted on the
 > >  > way out, too.
 > >  >
 > >  > Just to make troubleshooting difficult, this problem only shows up
 > >  > after the system has been up for roughly 36 hours, depending on the
 > >  > amount of traffic.
 > >  >
 > >  > I am using the latest bios that I am aware of. The bios that I
 > >  > recently applied did include a firmware update for the realtek
 > >  > interface, but this did not affect the problem.
 > >  > _______________________________________________
 > >
 > >  I disabled some hw features and works fine.
 > >  like this (/etc/rc.conf)
 > >  ifconfig_re0="inet 192.168.1.10  netmask 255.255.255.0 -rxcsum -txcsum
 > >  -tso -lr
 > >  o"
 > >
 > >  /Eric
 > >
 > 
 > I experienced the same problems shortly after upgrading to 7.0-PRE
 > After about a day and something of uptime my ssh shells began to drop
 > with messages about corrupted/mismatched checksums.
 > I "fixed" the problem by putting an em(4) interface in the machine,
 > because I need it to be online and accessible remotely at all times.
 > 

There had been several bus_dma(9) related bugs in re(4) for a long
time and I guess I fixed most of them. Hiding actual bugs by
replacing interfaces is not a good way to fix root cause of the issue.
If you can reproduce above issues in HEAD please let me know.

-- 
Regards,
Pyun YongHyeon