the freebsd em driver periodically hangs when running in a VM on all versions of esxi after (but not including) build 799733 (ESXi 5.0 GA). The symptoms of the problem are that the freebsd vm drops traffic on the affected interface for a period of time. The kernel reports errors: Apr 22 19:12:08 xxxx kernel: em0: Watchdog timeout -- resetting Apr 25 07:38:18 xxxx kernel: em0: Watchdog timeout -- resetting Apr 27 23:29:35 xxxx kernel: em1: Watchdog timeout -- resetting Apr 29 18:03:29 xxxx kernel: em1: Watchdog timeout -- resetting May 5 05:28:38 xxxx kernel: em1: Watchdog timeout -- resetting FreeBSD 8/9 were particularly badly affected by this, to the point that the em driver was totally unusable on those versions when running as vm after esxi build 799733, i.e ~30% packet loss for pings. The intel em driver update in 10.0 helped quite a bit, and anecdotally the problem seems to be less bad on ESXi 6.0. Now the figure seems to be down at around 1-2% pktloss over extended periods of time. However it's still there on 10.1 with all recent versions of esxi, and prevents freebsd from being able to function properly as a router when using the em interface. The problem also affects freebsd derivatives (e.g. some versions of JunOS Firefly / vRR / vMX, etc).
^Triage: correct assignment. Discussed with: koobs@.
I realize that this bug is very old. Does the bug still occur? If so, which virtual NIC is used? Can a vmware.log. vmx configuration file and vmkernel.log be provided? Thanks!
sorry, haven't checked in some years and I don't have an esxi installation where I can easily install freebsd to check it. This specific problem affected the freebsd em driver only, not the vmx driver (which had a different set of problems on the esxi platform - see bug #198406)