Hello. Stuart i have 8 OpenBSD routers with em(4) and OpenBSD 5.2 (MSI enabled). It seems some of our SMTP(s) connections (with attachments) are unstable but it's very very random (~1/500). Other protocols are more stable but a little slower due to errors.
Here are my stats on Intel i350 servers (with LACP trunks and not): remote BGP router 1 Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Colls em0 1500 <Link> a0:36:9f:XX:XX:ac 1333571 0 12169009 0 0 em1 1500 <Link> a0:36:9f:XX:XX:14 145522555 127 170665528 0 0 em2 1500 <Link> a0:36:9f:XX:XX:14 279931109 214 213998835 0 0 em3 1500 <Link> a0:36:9f:XX:XX:ac 92556655 201 130725384 0 0 remote BGP router 2 Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Colls em0 1500 <Link> a0:36:9f:XX:XX:1e 6669245 11 188865 0 0 em1 1500 <Link> a0:36:9f:XX:XX:cc 46412467 0 36851370 0 0 em2 1500 <Link> a0:36:9f:XX:XX:cc 46853863 0 99470829 0 0 em3 1500 <Link> a0:36:9f:XX:XX:1e 78433193 294 3776291 0 0 remote client router 1 Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Colls em0 1500 <Link> a0:36:9f:XX:XX:32 57014143 0 44354406 0 0 em1 1500 <Link> a0:36:9f:XX:XX:6c 34387984 142510 89895370 0 0 em2 1500 <Link> a0:36:9f:XX:XX:6c 45645663 0 65272767 0 0 em3 1500 <Link> a0:36:9f:XX:XX:32 48745525 0 33779694 0 0 local client router 1 Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Colls em0 1500 <Link> 68:05:ca:XX:XX:e0 1679516672 35681 1572183435 0 0 em1 1500 <Link> 68:05:ca:XX:XX:e1 1908150361 121884 2196221765 0 0 Here is a dmesg example (same on all servers, all are Dell R320 with bge disabled on bios and OpenBSD) em0 at pci2 dev 0 function 0 "Intel I350" rev 0x01: msi, address a0:36:9f:XX:XX:1e em1 at pci2 dev 0 function 1 "Intel I350" rev 0x01: msi, address a0:36:9f:XX:XX:1f If it can help OpenBSD team. Please not i cannot update to 5.3 because of big problems on OpenBSD on my Dell R320 (system freezes). I'm waiting OpenBSD 5.4 and less clients to test it on a slave server. -- Best regards, Loïc BLOT, UNIX systems, security and network engineer http://www.unix-experience.fr Le mardi 01 octobre 2013 à 08:37 +0000, Stuart Henderson a écrit : > On 2013-10-01, Patrick Lamaiziere <patf...@davenulle.org> wrote: > > Hello, > > > > With OpenBSD 5.3, our firewall does not handle our network load well. > > We loose around 5% of packets and netstat shows a lot of Ierr. > > > > That worked much better with 5.1. There was a change to not enable MSI > > on 82572 chipset on our Intel card ( "Intel PRO/1000 QP (82571EB)" rev 0x06) in 5.2 : > > > > http://freshbsd.org/commit/openbsd/a47ca448720823019bc6c618bf178a47fd1af73a > > > > My question is: could it be the cause of our load problem ? > > > > 5.1: > > em0 at pci5 dev 0 function 0 "Intel PRO/1000 QP (82571EB)" rev 0x06: msi, address 00:15:17:ed:98:9d > > em1 at pci5 dev 0 function 1 "Intel PRO/1000 QP (82571EB)" rev 0x06: msi, address 00:15:17:ed:98:9c > > em2 at pci6 dev 0 function 0 "Intel PRO/1000 QP (82571EB)" rev 0x06: msi, address 00:15:17:ed:98:9f > > em3 at pci6 dev 0 function 1 "Intel PRO/1000 QP (82571EB)" rev 0x06: msi, address 00:15:17:ed:98:9e > > > > 5.3 (on another box with the same hardware): > > em0 at pci5 dev 0 function 0 "Intel PRO/1000 QP (82571EB)" rev 0x06: apic 1 int 13, address 00:15:17:ed:98:65 > > em1 at pci5 dev 0 function 1 "Intel PRO/1000 QP (82571EB)" rev 0x06: apic 1 int 6, address 00:15:17:ed:98:64 > > em2 at pci6 dev 0 function 0 "Intel PRO/1000 QP (82571EB)" rev 0x06: apic 1 int 15, address 00:15:17:ed:98:67 > > em3 at pci6 dev 0 function 1 "Intel PRO/1000 QP (82571EB)" rev 0x06: apic 1 int 13, address 00:15:17:ed:98:66 > > > > We don't have any problem with this card, how can we re-enable MSI (without reverting this change)? > > Simplest way to test is to just revert that change in your source tree.. > That will identify whether this issue is due to disabling MSI, or whether > it's due to one of the many other changes between 5.1 and 5.3.. [demime 1.01d removed an attachment of type application/pgp-signature which had a name of signature.asc]